THESE DE DOCTORAT DE

ONIRIS COMUE UNIVERSITE BRETAGNE LOIRE

ECOLE DOCTORALE N° 605 Biologie Santé Spécialité : Santé Publique

Par Mariane POURCHET

Development of non-targeted approaches to evidence emerging chemical hazard

Identification of new biomarkers of internal human exposure, in order to support human biomonitoring and the study of the link between chemical exposure and human health

Thèse présentée et soutenue à Nantes, le 8 octobre 2020 Unité de recherche : LABERCA UMR INRAE 1329

Rapporteurs avant soutenance :

Benedikt WARTH Associate professor, University of Vienna, Vienna, Austria Katrin VORKAMP Doctor, Aarhus University, Aarhus, Denmark Development of non-targeted approaches to evidence emerging Composition du Jury : chemical hazard Président : Adrian COVACI Professor, University of Antwerp, Antwerp, Belgium

ExaminateursIdentification : Benedikt of WARTHnew biomarkers Associateof internal professor, human University exposure, of Vienna, in Vienna, order Austria to support Katrin VORKAMP Doctor, Aarhus University, Aarhus, Denmark human biomonitoringJana KLANOVA and the studyProfessor, of the Masaryk link University,between Brno, chemical Czech Republic exposure and Adrian COVACI Professor, University of Antwerp, Antwerp, Belgium human health Dir. de thèse : Jean-Philippe ANTIGNAC Doctor, Oniris, Nantes, France Thèse présentée et soutenue à Nantes, le 8 octobre 2020 Unité de recherche : LABERCA UMR INRAE 1329

I would like to express my sincere gratitude to my thesis committee,

Benedikt Warth, Associate professor, Department of Food Chemistry and Toxicology, University of Vienna, Austria

Katrin Vorkamp, Senior researcher, Department of Environmental Science, Aarhus University, Denmark

Adrian Covaci, Professor, Toxicological Centre, University of Antwerp, Belgium

Jana Klánová Professor, Research Centre for Toxic Compounds in the Environment (RECETOX), Masaryk University, Brno, Czech Republic

for their time in reading and evaluating this manuscript.

Les présents travaux de thèse réalisés au LABERCA m’ont amenée à faire de nombreuses rencontres. Je tiens à exprimer ma sincère gratitude et à remercier l’ensemble des acteurs qui m’ont accompagnée pendant ces trois années.

Je tiens à remercier sincèrement Bruno Le Bizec, directeur du LABERCA, pour m’avoir accueillie dans un environnement scientifiquement et humainement propice au travail, ainsi que pour votre regard critique et bienveillant sur mon travail.

Un grand merci à Jean-Philippe Antignac, directeur de thèse, pour m’avoir accompagnée pendant ces trois années. Merci également pour l’autonomie que tu m’as accordée et pour nos nombreuses discussions qui m’ont permises d’enrichir mes connaissances aussi bien en sciences qu’en gestion de projet.

Aux membres de mon comité de suivi individuel, Laurent Debrauwer et Jean-Charles Martin, je vous remercie pour le temps que vous avez consacré à évaluer en toute bienveillance mes travaux ainsi que pour vos précieux conseils.

I want to express my gratitude to all HBM4EU partners for all the interesting discussions we had during meetings and conferences. I also want to thank Marja Lamoree, Jeroen Meijer and the team members of the VU-E&H, as well as Jana Klanova, Elliott Price and the team members of RECETOX for the wonderful secondments I had in Amsterdam and Brno and for sharing their knowledge. I want to thank the University of Granada and the University of Masaryk for sending placenta and meconium samples.

Merci à Ronan Cariou, pour m’avoir accompagnée dans le laboratoire, ton enthousiasme et ta rigueur. Je souhaite un grand succès à HaloSeeker !

Merci à Emmanuelle Bichon, pour m’avoir guidée sur le développement de mes méthodes et pour m’avoir écoutée. Je souhaite une longue vie au CCOA !

Merci à Ingrid Guiffard, ma maman de labo, pour m’avoir formée à l’utilisation de la GPC, guidée sur le choix des standards et surtout pour ton soutien sans faille.

Merci à Gaud Dervilly, pour m’avoir accompagnée depuis notre discussion à la fin d’un cours de master jusqu’à l’obtention d’un poste post-thèse

Merci à Yann Guitton et Sébastien Hutinet, pour le développement des outils informatiques, votre disponibilité et l’effort conjoint pour que l’on parle la même langue.

Merci aux membres de la plateforme, Fabrice Monteau, Karinne Pouponneau, Marie- Line Morvan, Stéphanie Prévost et Sadia Ouzia, pour m’avoir formée sur les instruments, aidée à trouver mes repères dans le laboratoire et initiée à Galaxy. Merci à Anne-Lise Royer, pour tes conseils en métabolomique, ta bonne humeur ainsi que tous les bons moments passés ensemble.

Merci à l’ensemble de l’équipe UCO, pour le partage de connaissances, l’enregistrement de standards et surtout pour votre aide sur les expériences ciblées.

Merci aux membres UPC et ATI pour le partage du matériel, de votre temps et de vos connaissances.

Merci à Florence Ramdin, Lucie Besecque et Virginie Klingler pour votre aide dans les différentes tâches administratives, avec parfois quelques nœuds au cerveau pour gérer les budgets extérieurs. Merci pour votre patience et votre réactivité.

Cette thèse n’aurait pas été la même sans l’équipe de Docs & Co, à l’intérieur comme à l’extérieur du bocal. Je tiens à remercier l’ensemble des doctorants et post-doctorants passés ou encore présents, pour la perpétuelle bonne humeur et pour l’esprit familial que vous avez créé. Malgré un épisode Covid non propice aux interactions sociales, je garde de très bons souvenirs du club fromage, du Monday cake et des repas de Noël, des mots-fléchés, des photos de groupes et du calendrier de l’Avent. Merci à Marie, ma sœur de thèse et confidente, pour nos discussions entre nos deux bureaux, ton soutien et tous les bons moments passés ensemble ; à mes mamans de bureau Léa et Elsa, toujours de bonne humeur et prêtes à partager les derniers potins ; Maykel, mon grand frère de labo, pour ta grande gentillesse, ton soutien et les cours d’espagnol ; Caroline, pour ta positive-attitude en toute circonstance et ton soutien (merci également pour la relecture du manuscrit et la valorisation de mon travail) ; Jérémy, pour tes blagues, imitations et références filmographiques toujours bien dosées (merci aussi pour la correction de l’anglais du manuscrit, j’ai beaucoup ri en lisant certains commentaires). Merci à Manon, pour m’avoir motivée à aller aux séances de sport quand il pleuvait ; Luca, pour avoir pris le temps de m’apprendre les bases des modèles OPLS et pour les succulentes spécialités italiennes que tu nous as fait (re)découvrir (mozza, noisettes, fromages, charcuterie) ; Julien, ou devrais-je dire Satan, mais avec un cœur en guimauve ; Gabriel, le cycliste engagé, avec un

jeu de mots pour chaque situation ; Komodo, pour ton ouverture d’esprit ; Ahmed, l’incarnation de la sagesse et de la gentillesse ; Alexis, pour m’avoir guidée sur mes premières manips et pour ton grand cœur ; Inas, dont la générosité dépasse les limites du réel ; Chloé, pour ton dynamisme et ta diarrhée orale ; Mikaïl, le grand gaillard musclé qui ne ferait pas de mal à une petite mouche. Thomas et Yoann, les inséparables ou presque, toujours le sourire aux lèvres et prêts à blaguer. Merci également à Annabelle et Lucie, pour avoir tenté l’expérience d’un stage sous ma tutelle, je vous souhaite de vous épanouir dans votre carrière.

A tous mes amis en dehors du laboratoire, je vous remercie pour les merveilleux moments que je passe en votre compagnie et pour la curiosité que vous avez apportée à ma thèse, surtout autour de ce fameux « cacanium ».

Un énorme merci à ma famille, ma petite sœur Malaurie, mon oncle Manou et principalement à ma maman qui a toujours eu confiance en moi et sans qui je ne serai pas la personne que je suis aujourd’hui. Merci à mon compagnon Raphaël, pour me cuisiner des bons petits plats chaque jour et pour l’amour dont tu me gâtes à chaque instant.

TABLE OF CONTENTS

Table of contents

TABLE OF CONTENTS ...... 11

RESUME DE LA THESE EN FRANCAIS ...... 17

GENERAL INTRODUCTION ...... 27

CHAPTER 1 ...... 33

NON-TARGETED SCREENING OF BIOMARKERS OF HUMAN INTERNAL CHEMICAL EXPOSURE: STATE-OF-THE- ART AND CHALLENGES ...... 33

1.1. HUMAN CHEMICAL EXPOSURE: CONTEXT AND CHALLENGES ...... 35 1.1.1. The Exposome concept ...... 35 1.1.2. Chemical contaminants ...... 35 1.1.3. Routes of external human exposure to chemical contaminants ...... 37 1.1.4. Internal exposure and Absorption, Distribution, Metabolism and Excretion (ADME) processes ...... 38 1.1.5. Focus on early life stage of exposure ...... 41

1.2. HUMAN BIOMONITORING AND THE HBM4EU PROJECT ...... 41 1.2.1. Human biomonitoring programmes ...... 41 1.2.2. Human Biomonitoring for Europe (HBM4EU) ...... 43 1.2.2.1. Global overview ...... 43 1.2.2.2. Work-package 16 “emerging chemicals” ...... 44 1.3. DEFINITIONS FOR SETTING THE SCENE ...... 45 1.3.1. Targeted screening ...... 46 1.3.2. Suspect screening ...... 47 1.3.3. Non-targeted screening ...... 47

1.4. CHALLENGES OF NON-TARGETED SCREENING WORKFLOWS ...... 48 1.4.1. Preliminary analytical background ...... 48 1.4.2. Sample preparation ...... 50 1.4.2.1. The selectivity versus sensitivity compromise ...... 50 1.4.2.2. The starting sample volume compromise ...... 51 1.4.2.3. Extraction methods ...... 52 1.4.2.4. Additional purification and fractionation ...... 54 1.4.2.5. Extract reconstitution ...... 55 1.4.3. Analysis ...... 55 1.4.3.1. Chromatographic separation ...... 56 1.4.3.2. Ionisation and detection ...... 58 1.4.3.3. Effect-directed analysis ...... 62 1.4.4. Data processing ...... 64 1.4.4.1. Post-acquisition data pre-processing ...... 64 1.4.4.2. Compound annotation and identification ...... 65 1.4.5. Method performance assessment ...... 67

Table of contents

1.5. CONCLUSION ...... 70

CHAPTER 2 ...... 73

INSTRUMENTAL METHOD DEVELOPMENT AND DATA PROCESSING TOOLS ...... 73

2.1. INTRODUCTION ...... 75

2.2. INSTRUMENTAL METHOD OPTIMISATION ...... 76 2.2.1. QA/QC mix of standards ...... 76 2.2.2. LC-HRMS method development ...... 80 2.2.2.1. LC method development ...... 80 2.2.2.2. MS method development ...... 86 2.2.3. GC-HRMS method development ...... 90 2.2.3.1. GC method development ...... 90 2.2.3.2. Conclusion on GC-MS method development...... 93 2.3. LC- AND GC-HRMS METHOD PERFORMANCE ...... 95 2.3.1. Calibration curve ...... 95 2.3.2. Instrumental sensitivity ...... 96 2.3.3. Instrument repeatability ...... 98 2.3.4. Discussion ...... 99

2.4. DATA PROCESSING AND BIOINFORMATICS TOOLS DEVELOPED FOR NON-TARGETED SCREENING ...... 100 2.4.1. Recently developed software to process non-targeted data generated by HRMS instruments ...... 100 2.4.1.1. Methodology to screen for halogenated compounds by HRMS ...... 101 2.4.1.2. HaloSeeker software ...... 102 2.4.1.3. MS-DIAL software ...... 105 2.4.1.4. HBM4EU database ...... 106 2.4.2. Strategy developed to process data generated by non-targeted approaches ...... 107 2.4.2.1. Investigation of LC-HRMS data ...... 107 2.4.2.2. Creation of LC-HRMS database ...... 112 2.4.2.3. Investigation of GC-HRMS data ...... 113 2.5. MANAGEMENT OF THE EXTERNAL CONTAMINATION ...... 115

2.6. CONCLUSION ...... 116

CHAPTER 3 ...... 119

DEVELOPMENT OF NON-TARGETED SAMPLE PREPARATION PROTOCOLS TO SCREEN HALOGENATED MARKERS OF CHEMICAL EXPOSURE IN HUMAN SAMPLES ...... 119

3.1. INTRODUCTION ...... 121

3.2. ACIDIC HYDROLYSIS AND GPC: APPLICATION TO ADIPOSE TISSUE ...... 123 3.2.1. Introduction ...... 123 3.2.2. Acidic hydrolysis protocol ...... 124 3.2.2.1. Sample preparation and analysis ...... 124

Table of contents

3.2.2.2. Matrix effect ...... 124 3.2.3. GPC protocol ...... 128 3.2.3.1. Sample preparation ...... 128 3.2.3.2. Instrumental method ...... 128 3.2.3.3. Comparison of GPC with acidic hydrolysis ...... 131 3.2.3.4. Investigation of external contamination by GPC ...... 132 3.2.4. Discussion and conclusion ...... 136

3.3. BLIGH AND DYER METHOD: APPLICATION ON PLACENTA ...... 137 3.3.1. Introduction ...... 137 3.3.2. Method development ...... 137 3.3.2.1. Protocol ...... 137 3.3.2.2. Method performance ...... 138 3.3.2.3. Comparison with targeted method ...... 142 3.3.2.4. Sample amount for analysis ...... 146 3.3.3. Discussion and conclusion ...... 147

3.4. CAPTIVA EMR-LIPID®: APPLICATION TO HUMAN MILK AND MECONIUM ...... 148 3.4.1. Introduction ...... 148 3.4.2. Method development ...... 149 3.4.2.1. Protocol adaptation to human milk ...... 149 3.4.2.2. Sample volume for analysis ...... 155 3.4.2.3. Protocol transposition to meconium samples ...... 158 3.4.2.4. Sample amount for analysis ...... 161 3.4.2.5. Additional rinsing step ...... 163 3.4.2.6. Liquid-liquid partitioning optimisation ...... 165 3.4.3. Method assessment ...... 167 3.4.3.1. Linearity ...... 167 3.4.3.2. Repeatability ...... 168 3.4.3.3. Recovery ...... 168 3.4.3.4. Matrix effect ...... 168 3.4.3.5. Predicting the recovery through the physicochemical properties of the considered markers ...... 173 3.4.3.6. Comparison with targeted method ...... 178 3.4.4. Compatibility with EDA approach ...... 179 3.4.4.1. Partnership context ...... 179 3.4.4.2. EDA protocol ...... 180 3.4.4.3. Whole extract bioassay analysis ...... 181 3.4.4.4. Compatibility of the Captiva EMR-Lipid® sample preparation with EDA ...... 184 3.4.5. Discussion and conclusion ...... 187

3.5. CONCLUSION ...... 189

CHAPTER 4 ...... 191

Table of contents

CHARACTERISATION OF INTERNAL HUMAN EXPOSURE TO HALOGENATED CONTAMINANTS IN HUMAN SAMPLES BY NON-TARGETED SCREENING ...... 191

4.1. INTRODUCTION ...... 193

4.2. ANALYSIS OF ADIPOSE TISSUE SAMPLES ...... 193 4.2.1. Introduction ...... 193 4.2.2. Adipose tissue sample investigation ...... 194 4.2.3. Conclusion ...... 197

4.3. ANALYSIS OF PLACENTA SAMPLES ...... 197 4.3.1. WP14-WP16 interaction ...... 197 4.3.2. Placenta sample investigation ...... 198 4.3.3. Conclusion ...... 201

4.4. ANALYSIS OF HUMAN MILK SAMPLES...... 201 4.4.1. Samples and protocol ...... 201 4.4.2. Human milk sample investigation ...... 202 4.4.2.1. Investigation of LC-HRMS data ...... 202 4.4.2.2. Investigation of GC-HRMS data ...... 206 4.4.3. Conclusion ...... 208

4.5. ANALYSIS OF MECONIUM SAMPLES ...... 209 4.5.1. Samples and protocol ...... 209 4.5.2. Meconium sample investigation ...... 210 4.5.2.1. Investigation of LC-HRMS data ...... 210 4.5.2.2. Investigation of GC-HRMS data ...... 211 4.5.3. Conclusion ...... 213

4.6. CONCLUSION ...... 213

GENERAL CONCLUSION AND PERSPECTIVES ...... 215

LIST OF BIBLIOGRAPHICAL REFERENCES ...... 223

LIST OF ABBREVIATIONS ...... 245

LIST OF FIGURES ...... 251

LIST OF TABLES ...... 259

LIST OF SCIENTIFIC VALORISATIONS ...... 263

APPENDICES ...... 269

RESUME DE LA THESE EN FRANCAIS

- 17/296 -

Résumé de la thèse en francais

L'exposition aux substances chimiques est un sujet au centre de nombreuses préoccupations à différents niveaux. Les citoyens témoignent tout d’abord d’une prise de conscience grandissante du risque associé à l'exposition à ces substances pour leur propre santé, celle de leurs descendants et pour la biodiversité en général. L'impact toxicologique de certaines de ces substances chimiques et des mélanges qui leur sont associés sur la santé humaine, ainsi que la réalité des niveaux d'exposition pour la population en général et pour des sous-populations particulières, sont ensuite des questions traitées par les autorités compétentes, c'est-à-dire les gestionnaires du risque. Les mesures réglementaires destinées à protéger la santé humaine sont évalués par des programmes de biosurveillance (Louro et al., 2019) qui visent à générer des données précises sur l'exposition pour l'ensemble ou des sous-groupes particuliers de la population (en fonction du sexe, de l'âge, de la zone géographique, de l'appartenance ethnique, de la classe sociale, etc.).

La nécessité de documenter la réalité de cette exposition chimique humaine et de son effet sur la santé fait partie d'un nouveau domaine appelé "exposomique". Le concept d'exposome a été introduit et défini comme l'ensemble des expositions rencontrées au cours d'une vie, à partir de la période fœtale, ainsi que les effets biologiques associés (Wild, 2005). L'étendue de cet exposome global apparait ainsi extrêmement large, englobant l'exposition chimique (pesticides, plastifiants, etc.), biologique (virus, bactéries, etc.) et physique (ondes, bruit, etc.) et plus généralement les facteurs environnementaux liés aux modes de vie.

L'exposome chimique se réfère à la présence de contaminants chimiques dans le continuum environnement-alimentation-santé. Ce dernier est une entité dynamique qui évolue constamment et diffère selon les zones géographiques (par exemple, différentes activités anthropiques, la météo, etc.). Sa caractérisation est donc complexe et représente un travail titanesque impliquant des efforts collaboratifs au niveau international. La recherche de marqueurs d'exposition directement dans le corps humain (fluides et tissus biologiques) correspond à l'exposition chimique interne. Les méthodes ciblées sont couramment utilisées pour rechercher de manière quantitative des contaminants connus (par exemple, dioxines, PCB, certains pesticides). Cependant, ces approches sont limitées à un ensemble de composés et ne permettent pas de détecter les composés d’intérêt émergent (CEC) (Sauvé et Desrosiers, 2014). De nouvelles approches dites non-ciblées sont donc en cours de développement pour caractériser de manière exhaustive des échantillons sans connaissance au préalable (Sobus et al., 2019). Les approches non-ciblées ont d'abord été développées pour les études

- 19/296 - Résumé de la thèse en francais environnementales et alimentaires, alors qu'elles sont encore en train d'émerger pour les études d'exposition interne chez l’Homme. Ces approches sont aujourd'hui favorisées par la mise en œuvre croissante de technologies avancées telles que la chromatographie liquide ou en phase gazeuse couplée la spectrométrie de masse à haute résolution (LC/GC-HRMS). En revanche, ces approches sont également confrontées à un certain nombre de défis en termes de développement, d'optimisation et d'harmonisation (Pourchet et al., 2020). Les difficultés majeures résident dans i) le bon compromis à trouver pour atteindre une sensibilité suffisante tout en détectant un large éventail de molécules, ii) l’annotations et l’identifications des composés avec les logiciels et bases de données en cours de développement, iii) ainsi que l’ensemble des défis liés aux matrices humaines (faibles volumes disponibles, faibles niveaux de contamination).

Au niveau européen et dans le cadre de l'initiative Horizon 2020, le projet HBM4EU (www.hbm4eu.eu) de biosurveillance humaine pour l'Europe a été lancé en 2017 (Ganzleben et al., 2017). Un budget de 70 millions d'euros réparti sur une centaine laboratoires partenaires situés dans trente pays européens a été alloué sur les cinq ans. Cet effort conjoint vise à harmoniser les approches de génération de données sur l'exposition entre les États membres afin d'agir plus efficacement sur la protection de la santé des citoyens. Tous les partenaires collaborent pour caractériser l'exposition aux produits chimiques en Europe et ainsi établir la relation entre l'environnement et la santé humaine. Ces objectifs ambitieux du projet HBM4EU sont répartis sur un total de seize groupe de travail (WP), allant du travail de laboratoire tel que le développement de méthodes analytiques et la génération de données, aux discussions socio- politiques avec les parties prenantes européennes. Une composante de ce projet, le WP16 "substances chimiques émergentes", est spécifiquement consacrée au développement, à l'application et à l'harmonisation d'approches non-ciblées. Il vise à stimuler le développement de ces nouvelles stratégies pour caractériser l'exposition humaine interne afin de découvrir de nouveaux marqueurs d'exposition.

Dans le cadre du présent travail de doctorat, inclus dans le WP16 du projet HBM4EU, tous les défis analytiques susmentionnés associés au développement d’approches non-ciblées pour la caractérisation de l'exposition humaine aux contaminants ont été abordés. Des matrices complexes telles que le tissu adipeux, le placenta, le lait maternel et le méconium, reconnues comme des compartiments de stockage et/ou d'excrétion pour les contaminants chimiques, ont été étudiées. En outre, elles sont représentatives de l'exposition périnatale et peuvent être

- 20/296 - Résumé de la thèse en francais caractéristiques d'une exposition à un stade précoce de la vie, qui peut affecter la santé pendant les stades de développement. En effet, un certain nombre d'études ont déjà montré que, pendant cette période de la vie, certains contaminants peuvent avoir un impact néfaste (Maccari et al., 2014).

Compte tenu de l'étendue extrêmement vaste des travaux sous-jacents à effectuer pour traiter cette thématique, la priorité a été donnée à la détection des contaminants chlorés et bromés en raison de la toxicité déjà reconnue pour certaines classes de substances (par exemple PCB, BFR, pesticides organochlorés). Les principaux objectifs suivants ont été fixés dans le cadre de ce travail de thèse : i) développer des approches non-ciblées pour la caractérisation de l'exposition interne de l'Homme, avec un accent particulier sur la période périnatale et l'exposition aux contaminants organiques halogénés, ii) identifier les principaux défis qui doivent être relevés pour le développement de ces méthodes afin de guider les laboratoires souhaitant mettre en œuvre ces approches dans leurs activités et iii) mettre en exergue les perspectives les plus prometteuses permettant de consolider et d’harmoniser le développement des approches non- ciblées. Le présent travail de doctorat a été cofinancé par le projet HBM4EU et la région des Pays de la Loire (France). Il a intégré des collaborations externes avec l'Université d'Amsterdam et de l'Université de Masaryk, dans le cadre des labels européen et Agreenium.

Les différentes étapes du flux de travail analytique non-ciblé, de la préparation des échantillons à l'analyse et au traitement des données, ont été étudiées. Diverses matrices humaines, dont le tissu adipeux, le placenta, le lait humain et le méconium, ont été sélectionnées afin de caractériser plus spécifiquement l'exposition interne de l'Homme aux substances chimiques à un stade précoce de la vie (pendant la période périnatale). Le principal défi consistait alors à saisir un large éventail de marqueurs d'exposition présentant diverses propriétés physico- chimiques avec une sensibilité et une sélectivité suffisantes. L'élimination suffisante des interférences matricielles (par exemple les protéines, les lipides), ainsi que l'utilisation de faibles volumes d'échantillons communément disponibles pour les matrices humaines (c'est-à- dire de 100 mg à 1 g) ont constitué les principaux facteurs de contrainte pris en compte lors de ce développement. Différents protocoles de préparation d'échantillons ont été évalués, à savoir l'hydrolyse acide, le fractionnement par chromatographie sur gel (GPC), la séparation par Bligh et Dyer et l'extraction en phase solide sur Captiva EMR-Lipid®. Toutes les options de préparation des échantillons testées présentaient des avantages et des limites qui ont été identifiés et discutés. Enfin, les méthodes de fractionnement des échantillons, telles que la GPC,

- 21/296 - Résumé de la thèse en francais sont apparues particulièrement prometteuses, car elles sont conformes à la philosophie globale des analyses non-ciblées, compte tenu de leur caractère exhaustif et préservant l'échantillon analysé. Les protocoles basés sur l'élimination des composés endogènes, tel que la cartouche Captiva EMR-Lipid®, sont également apparus comme des options pertinentes pour les analyses non-ciblées. Les deux approches semblent intéressantes et devraient davantage être étudiées. Elles peuvent également être couplées à d'autres préparations d'échantillons plus sélectives. En effet, le phénomène d’effet matrice est un problème majeur lors du développement de préparation d’échantillon non-ciblée, bien plus préoccupant que pour le développement de protocoles ciblés.

De plus, les protocoles de préparation d'échantillons développés doivent être adaptés en fonction de l'instrumentation utilisée. Pour la présente étude, un partage liquide-liquide a été effectué et a donné lieu à des fractions polaires et non polaires analysées par LC-ESI(+/-) et GC-EI-HRMS, respectivement. Ces méthodes instrumentales complémentaires ont été développées afin d'augmenter le nombre de marqueurs d'exposition accessibles. Ainsi, les paramètres génériques ont été optimisés et cette analyse complémentaire a permis d'étendre les gammes de masse (100-1000 Da) et de log P (de 2 à 8) des composés détectables. Par ailleurs, les détecteurs HRMS sont nécessaires pour atteindre une précision de masse élevée, indispensable à l'identification des composés. Dans le cas présent, la technologie Orbitrap a été utilisée, à la fois en mode de balayage complet pour une caractérisation large et en mode MS/MS (DDA itérative) pour l'identification. Pour la présente étude, la séparation chromatographique a été assurée dans une dimension, mais des dimensions supplémentaires telles que la LCxLC, la GCxGC, et/ou la spectrométrie de mobilité ionique représentent également des technologies complémentaires d'intérêt. À l'heure actuelle, la principale limite concernant ces approches analytiques avancées pourrait être la disponibilité de logiciels optimisés capables de traiter des données aussi complexes, ce qui entrave leur application à grande échelle.

Enfin, le traitement des données est probablement l'étape la plus difficile du processus, car il ne requiert pas seulement des connaissances et une expertise analytique, mais aussi des compétences avancées en bio-informatique. Le traitement des données rencontre encore des difficultés pour mettre en évidence les signaux d'intérêt potentiel (c'est-à-dire les informations utiles) à partir du vaste volume de données brutes généré par ces approches. Le logiciel HaloSeeker, développé au LABERCA a été utilisé pour se concentrer plus spécifiquement sur

- 22/296 - Résumé de la thèse en francais les substances chlorées et bromées. Au cours du présent doctorat, des outils supplémentaires pour identifier et gérer les signaux résultant d'une contamination externe ont également été mis en œuvre (par exemple, alignement, soustraction de blancs). Cela a permis, grâce à la stratégie élaborée, de concentrer l'étude sur les composés exogènes potentiels. De plus, les réseaux moléculaires, déjà mis en œuvre dans les protocoles de travail de la métabolomique, apparaissent comme un autre outil prometteur qui pourrait être utilisé pour filtrer les signaux d'intérêt, comme par exemple, en regroupant les signaux liés aux interférences de la matrice (par exemple les lipides) ou les marqueurs fonctionnellement liés.

Enfin, une fois un signal considéré comme intéressant, l'annotation des marqueurs d'exposition a été effectuée en utilisant la base de données de référence d'annotation HBM4EU. Cependant, cette base de données est en construction et, dans l'état actuel, manque encore d'informations (telles que les spectres MS/MS) pour confirmer sans ambiguïté l'identification de tous les marqueurs. Il est important de noter que d'autres approches telles que la DDA itérative doivent être consolidées pour générer efficacement des informations structurales sur les signaux non annotés. Finalement, l'identification sans ambiguïté sera possible si le standard analytique correspondant au marqueur considéré est disponible. Nous pouvons facilement imaginer que ce n'est pas le cas pour les composés inconnus, pour lesquels des recherches supplémentaires seraient nécessaires afin d'élucider leur structure.

En ce qui concerne l'application de la méthode non-ciblée développée, certains marqueurs d'exposition chimique ont été identifiés avec succès grâce à l'expertise acquise au cours du présent projet et à des collaborations pluridisciplinaires. α-HBCDD, le triclosan et le 4-hydroxychlorothalonil ont été identifiés par LC-HRMS dans le tissu adipeux, le placenta et le lait humain, respectivement, ainsi que le p,p'-DDE et le HCB ont été identifiés par GC-HRMS dans le lait humain et le méconium. Ces premières preuves de concept démontrent les points forts des approches non-ciblées pour rechercher les contaminants organiques dans les matrices humaines sans connaissance a priori. Elles illustrent également les difficultés rencontrées pour obtenir de tels résultats, car les quelques composés identifiés représentent probablement une part restreinte de tous les contaminants présents dans l'échantillon. L'étape suivante consiste à approfondir les données déjà acquises afin de poursuivre l’investigation des données et d'identification des marqueurs. En outre, il est important de continuer à générer de nouvelles données d'exposition sur des échantillons réels par ces approches car l'accumulation de ces données peut également contribuer à faire progresser l'identification des marqueurs d'intérêt.

- 23/296 - Résumé de la thèse en francais

Une fois ces données non-ciblées acquises, elles peuvent être stockées pendant des années et retraitées avec de nouveaux outils, ce qui représente également un point fort de ces approches.

Dans le cadre de la collaboration avec la Vrije Universiteit, une autre approche de détection de marqueurs potentiels d'exposition d'intérêt a été évaluée, basée sur des tests d'activité biologique. Cette approche EDA nous a permis de combiner l'approche chimique développée ci-dessus avec un test biologique axé sur les perturbateurs du système hormonal thyroïdien. Les approches biologiques sont également apparues particulièrement intéressantes pour caractériser l'effet de mélange. Toutefois, la sensibilité de cette approche n'est pas encore compatible avec les niveaux de contamination attendus dans les échantillons humains. D'autres développements sont donc nécessaires. Cette approche complémentaire de l'approche non-ciblée développée au cours du présent travail de doctorat illustre le besoin de collaboration entre les laboratoires. En effet, l'EDA nécessite des équipements, des connaissances et des compétences spécifiques en biologie et en toxicologie, qui ne sont pas directement implémentables dans tous les laboratoires d'analyse.

L'évaluation des performances des méthodes non-ciblées a été un autre élément important du présent travail de doctorat. Le phénomène de l'effet de matrice a été étudié avec des approches qualitatives et quantitatives et nous conseillons d'inclure systématiquement cette expérience dans le processus de développement des méthodes non-ciblées. Une double caractérisation des mêmes échantillons avec une analyse ciblée et non-ciblée est également apparue intéressante au cours du développement des approches non-ciblées, afin i) de mieux caractériser l'écart existant entre les deux approches en termes de spécificité et de sensibilité et ii) de déterminer la raison pour laquelle un composé ne serait pas détecté entre la non-présence de celui-ci dans l’échantillon ou l’incapacité de la méthode à le détecter. Ensuite, la présente étude a confirmé le faible niveau de contamination des échantillons humains en ce qui concerne certains marqueurs d'exposition chimique, et la différence de sensibilité encore limitée entre les approches ciblées et non-ciblées. Il a été démontré que l'approche non-ciblée peut être actuellement utilisée pour caractériser les échantillons présentant les niveaux de concentration les plus élevés, c'est-à-dire les individus les plus exposés. D'une part, nos résultats globaux ont confirmé la pertinence et la compatibilité des méthodes non-ciblées et leur application à la biosurveillance humaine. D'autre part, ils ont démontré que des progrès supplémentaires sont encore nécessaires pour améliorer leur sensibilité et leur capacité de traitement.

- 24/296 - Résumé de la thèse en francais

La gestion de la contamination externe pouvant survenir lors du prélèvement, de la préparation et de l'analyse des échantillons est un autre aspect crucial des approches non-ciblées qui suscite encore de l'intérêt et pour lequel une harmonisation importante est nécessaire au niveau international. Par exemple, des récipients vierges pourraient être systématiquement inclus pendant la collecte des échantillons, comme cela a été établi en collaboration avec RECETOX pour la collecte d'échantillons de méconium. Comme exigence minimale, les blancs de procédure doivent être préparés et analysés en parallèle avec les échantillons de matrice. Ensuite, ces blancs doivent être étudiés plus en détail pour évaluer la répétabilité de la contamination externe et être ensuite capables de détecter les contaminants omniprésents.

En conclusion, nos résultats ont démontré que la stratégie analytique développée est efficace pour le dépistage non-ciblé des marqueurs de l'exposition chimique interne de l'Homme. Cette approche émergente suscite actuellement un intérêt extrêmement élevé, de grandes attentes et des développements techniques intenses au niveau international. Ces approches non-ciblées permettront de produire très rapidement une nouvelle génération de données étendues sur l'exposition interne et de contribuer à l'élargissement des connaissances sur l'exposition de l’Homme et d'offrir de nouvelles perspectives pour les études de santé environnementale. Au niveau réglementaire, dans le contexte de la biosurveillance humaine et sur le long terme, les approches non-ciblées peuvent représenter un nouvel outil d'alerte précoce pour soutenir les politiques et établir des priorités pour les programmes de biosurveillance et l'évaluation des risques. Toutefois, ce domaine émergent nécessite un effort important en termes d'harmonisation au niveau international pour anticiper une plus grande comparabilité des données et pour consolider l'interprétation des résultats. Il devrait être complémentaire à d'autres initiatives dans les domaines de l'environnement et de la sécurité alimentaire où ces approches sont également développées. Cela implique d'exporter ces méthodes non-ciblées sur l’ensemble des continents, y compris dans les pays où l'accès aux instruments et logiciels de dernière génération est réduit, ce qui compromet la mise en œuvre de la stratégie. Cela implique d'initier des collaborations internationales, à l'instar du projet HBM4EU et au-delà de l'Europe.

- 25/296 -

- 26/296 -

GENERAL INTRODUCTION

- 27/296 -

General introduction

Eye-catching labels such as “Bisphenol free”, "Pesticide free", "paraben free", "preservative free", among others, are increasingly appearing on our daily use manufactured and food products. They are elements of positive communication and act as selling points. Indeed, the chemical exposure to exogenous substances resulting from anthropogenic activities is a topic that sparks many conversations at different levels. Many citizens are becoming aware of the risk associated with the exposure to these substances for their own health, their descendants’ and for biodiversity in general. The toxicological impact of some of these chemicals and their associated mixture on human health, as well as the reality of the exposure levels for the general and particular sub-populations, are issues handled by the competent authorities, i.e. the risk managers. Numerous studies conducted over several years, involving different actors within the scientific and policy areas, are required to perform the appropriate risk assessment, then to document, support and evaluate the risk management actions. The concerned chemical agents deemed toxic can be either regulated or even banned. The outcomes of these regulatory measures designed to protect human health are then evaluated by biomonitoring programmes (Louro et al., 2019) that aim to generate accurate exposure data for the general or particular population subgroups (based on gender, age, geographical area, ethnicity, social class, etc.). In this way, trends over the years can be established and subsequently combined with epidemiologic approaches, thus enabling the determination of the relationship between environmental exposure and its impact on human health. Then, the generated data are used to support policymaking and regulations are adapted to these trends.

The need of documenting the reality of this human chemical exposure and its associated health effect is part of a new field called exposomics. The concept of exposome has been introduced and defined as the comprehensive set of exposures encountered over a lifetime, starting from the foetal period, as well as the related biological effects (Wild, 2005). Therefore, the extent of this global exposome appears extremely wide, encompassing chemical (pesticides, plasticizers, etc.), biological (viruses, bacteria, etc.) and physical (waves, noise, etc.) exposure and more generally lifestyle environmental factors. The chemical exposome refers more specifically to the presence of chemical contaminants in the environment-food-human continuum. This chemical exposome is a dynamic entity that constantly evolves and differs according to geographical areas (e.g. different anthropogenic activities, weather, etc.). Its characterisation is therefore complex and represents a titanic work involving the gathering of collaborative efforts at international level.

- 29/296 - General introduction

The generation of exposure data involves both the analysis of the living environment of individuals (food, textiles, air, cosmetics) and the search of markers of exposure directly in the human body (biological fluids and tissues). This is defined as external and internal exposure, respectively. The most widespread strategy applied to characterise the exposure is to search for known contaminants with methods specifically developed namely targeted methods. These approaches are then used to generate quantitative exposure data on known contaminants, some of which are already regulated (e.g. dioxin, PCBs, certain pesticides). However, they are limited to a set of compounds and do not allow the detection of compounds of emerging concern (CEC). The latter may be contaminants that have recently appeared in the environment (e.g. substitution products after compounds banning) or that have been present in the environment for a while but with emerging concern (Sauvé and Desrosiers, 2014). Characterisation of CEC also includes unknown compounds, which makes the characterisation of the chemical exposome even more complex and inaccessible for targeted methods. New so-called non-targeted approaches are therefore being developed. Their main objective is to comprehensively characterise samples without a priori knowledge (Sobus et al., 2019). These qualitative methods have a complementary role to the targeted methods, extending the global overview of the chemical exposure. In the long term, they would efficiently support risk management and policy-making in order to prevent the exposure of populations to new toxic contaminants through their early warning positioning. Non-targeted approaches were firstly developed for environmental and food studies, while they are still emerging for internal human exposure studies. These approaches are today promoted by the increasing implementation of advanced technologies such as liquid or gas phase chromatography coupled to high resolution mass spectrometry (LC/GC-HRMS) in analytical laboratories that open the door to large-scale and high throughput chemical profiling. On the other hand, these approaches are also facing a number of significant challenges in terms of development, optimisation and harmonisation are faced by scientists (Pourchet et al., 2020). The bottlenecks remain in the compromise to reach sufficient sensitivity while screening a wide range of molecules, the identification of the detected markers with software and database still under development, as well as the specific challenges related to human matrices (low volumes available, low levels of contamination).

At the European level and in the framework of the Horizon 2020 initiative, the Human Biomonitoring for Europe - HBM4EU (www.hbm4eu.eu) project was launched in 2017 and will continue until 2021 (Ganzleben et al., 2017). A budget of 70 million euros spread over more than one hundred partner laboratories located in thirty European countries has been

- 30/296 - General introduction allocated over the five years. This joint effort aims to harmonise approaches to generate exposure data between member states in order to act more efficiently on the protection of citizens' health. All the partners are working together to characterise the chemical exposure in Europe and thus establish the relationship between the environment and human health. These ambitious purposes of the HBM4EU project are spread over a total of sixteen work packages (WP), ranging from laboratory work such as the development of analytical methods and data generation, to socio-political discussions with European stakeholders. One component of this project, WP16 “emerging chemicals”, is specifically dedicated to the development, application and harmonisation of non-targeted approaches. It aims to stimulate the development of these new strategies to characterise the internal human exposure in order to expend the capabilities of exposure-health investigations, to discover new markers of exposure, to guide further prioritisation in terms of risk assessment, and finally to contribute to an early warning support to policy.

In the frame of the present Ph.D. work, included within the WP16 of the HBM4EU project, all of the aforementioned analytical challenges associated with the non-targeted screenings of human exposure to contaminants were addressed, through the investigation of complex matrices such as adipose tissue, placenta, human milk and meconium. These matrices are recognised as storage and/or excretion compartments for chemical contaminants. Furthermore, they are representative of the perinatal exposure, such exposure at early stage of life being recognised to be particularly critical with regard to human health at short and long term. Indeed, a number of studies have already shown that, during this particular life window, certain contaminants can have even more deleterious impact (Maccari et al., 2014), hence the interest in characterising this exposure.

Considering the extremely wide expanse of the underlying work to be performed to handle this thematic, a number of prioritisation and focusing criteria had to be made. In the context of this Ph.D., priority has been given to the detection of chlorinated and brominated contaminants because of the already proven toxicity of associated substance classes (e.g. PCBs, BFRs, organochlorine pesticides), and the following main objectives have been established: i) to develop non-targeted approaches for the characterisation of internal human exposure with a particular focus on the perinatal period and exposure to halogenated organic contaminants, ii) to identify the major challenges that have to be faced for the development of these methods in order to lay the foundations to guide laboratories wishing to implement these approaches in

- 31/296 - General introduction their activities and iii) to pave the way to the necessary directions for consolidating and harmonising this emerging field during the coming years. The present Ph.D. work was co- funded by the HBM4EU project and the Pays de la Loire region (France). It was conducted within the European and Agreenium labels, and have implied external collaboration at the University of Amsterdam and the University of Masaryk.

The present manuscript is dealing with the difference components of this Ph.D. work as follows:

Chapter 1 describes the state-of-the-art regarding the human chemical exposome and the non-targeted screening approaches applied for its characterisation. Related challenges already highlighted by the scientific community are detailed, as well as the positioning of this work with regard to the purposes of the HBM4EU project.

Chapter 2 details the development and optimisation of instrumental methods based on LC- and GC-HRMS analysis. The strategy established to process the generated raw data until the identification of markers of exposure is also presented. A focus has been made at this stage on the detection of halogenated markers of exposure.

Chapter 3 relies the development of non-targeted sample preparation of complex human matrices (adipose tissue, placenta, human milk and meconium). It includes the evaluation of different approaches and the comparison with targeted methods.

Chapter 4 is dedicated to the results obtained with the application of the developed strategies to the analysis of real samples for each matrix and their characterisation using the non-targeted methods presented in Chapter 2 and 3. The identification of markers of exposure is illustrated as well as perspective for further investigations.

Finally, all the scientific valorisation activities resulting from this Ph.D. work are listed in the last section of this manuscript.

- 32/296 -

1. CHAPTER 1

NON-TARGETED SCREENING OF BIOMARKERS OF HUMAN INTERNAL CHEMICAL EXPOSURE: STATE- OF-THE-ART AND CHALLENGES

- 33/296 -

Chapter 1. Non-targeted screening of biomarkers of human internal chemical exposure: state-of-the-art and challenges

1.1. Human chemical exposure: context and challenges

1.1.1. The Exposome concept

“Silent spring” book published in 1962 by Rachel Carson raised awareness about domino effect induced on biodiversity by the use of toxic chemicals, in particular phytosanitary products, and their possible impact on human health. This environmental chemical exposure factor is growing up in fields establishing the origin of non-transmissible and/or chronic diseases. Today, it appears as a major contributor to multifactorial diseases (together with environmental and genetic factors). Consequently, the characterisation and prevention of this environmental exposure appears nowadays as a major concern. It has been mainly addressed by epidemiologists using genome-wide association studies to explore association between risk factor and disease prevalence. However, this approach does not identify biomarkers of exposure or mixtures especially those interacting in the causal chain between exposure and health effect. In this context, Christopher Paul Wild introduced in 2005 the concept of Exposome to complete the genome-associated knowledge. It is defined as an entity “encompassing life-course environmental exposures from the prenatal period” (Wild, 2005). The exposome is then embracing all environmental and non-genetic factors (radiation, infection, life style, diet, pollution, etc.) that interact with human health through the entire life from conception to death. However, exact borders of the exposome area remains a matter of discussions at international level. Beyond this definition issue, a main component of the exposome has been described as including all chemicals interacting with the human body (Rappaport and Smith 2010). This component can be named the chemical exposome on which the present Ph.D. work has been focused.

1.1.2. Chemical contaminants

Chemical contaminants can be defined as all chemical substances unexpectedly found in the environment-food-human continuum. Those substances can be intentionally or unintentionally produced, then released in the environment (air, water, soil) and can have hazardous effects on human health. Contaminants can be classified in several ways, for instance according to their intended use, namely pesticides, flame retardants, pharmaceuticals, personal care products, plasticisers, etc. (Sauvé and Desrosiers, 2014). They can also be classified depending on their health effect such as endocrine disrupting chemicals (EDCs) that alter the hormonal system

- 35/296 - Chapter 1. Non-targeted screening of biomarkers of human internal chemical exposure: state-of-the-art and challenges

(WHO/PCS/EDC/02.2), for instance polychlorinated biphenyls (PCB) (Turyk et al., 2007). Some contaminants are also regulated based on their physicochemical properties as persistent organic pollutants (POPs) defined by United Nations Environment Programme (UNEP) in the Stockholm convention as persistent in the environment, bioaccumulative in fatty tissues of human and wildlife, transported over long distances and with harmful effect on the human health and the environment (UNEP Stockholm Convention on POPs).

Overall, contaminants are divided in two sub-groups named “legacy contaminants” and “chemicals of emerging concern (CEC)”. The former refers to well-known contaminants (sources, environmental fate, health effect) which have already been regulated or banned because of recognised hazardous adverse effect, such as some pesticides (lindane, DDT, etc.), dioxins and PCB, brominated and chlorinated flame retardants (polybrominated diphenyl ethers (PBDE), dechlorane, etc.), short chain chlorinated paraffins, etc. (Annex I and V of the European chemical agency – ECHA). Regarding CEC, and despite increasing societal, scientific and policy relevance, they are currently a matter of non-consensual definition and various terminologies (e.g. emerging chemicals, emerging substances, emerging contaminants…). Because the current definitions are not associated with a classification rationale, such as a common mode of action, property, intended use, or regulatory status, they maintain unclear usage and semantic confusion. As proposed by Sauvé and Desrosiers in 2014 (Sauvé and Desrosiers, 2014), CECs encompass both new compounds recently detected in the environment-food-human continuum (for instance, newly developed substitutes of banned and/or regulated chemicals) and compounds with known presence, yet for which concerns have recently increased (e.g. due to progress of analytical performance, newly identified sources, uses and/or routes of exposure, particularly exposed sub-population, toxicological evidence, evolution of regulatory dispositions…). Importantly, the biotransformation products (named as metabolites below) of CECs are included in this definition and are of particular relevance to biological matrices, including human samples. However, CEC metabolites are typically less known and studied compared to their parent compounds (Alves et al., 2014).

Such chemical contaminants are widely produced over the world (e.g. world pesticide production was estimated at almost 6 million tons in 2005 according to FAO data (Carvalho, 2017)) and have very different physicochemical properties (lipophilic/hydrophilic, polar/nonpolar, high/low volatility, organic/inorganic etc.). This extensive use combined with

- 36/296 - Chapter 1. Non-targeted screening of biomarkers of human internal chemical exposure: state-of-the-art and challenges a large scale of properties biologically compatible with human and wildlife increase the risk and routes of exposure.

1.1.3. Routes of external human exposure to chemical contaminants

The global picture of our exposure is a system interlaced of several routes of exposure in the biogeochemical cycle including biosphere, lithosphere, atmosphere and hydrosphere. The natural connexion and equilibrium between human, wildlife, soil, air and water is affected by anthropogenic activities (industry, agriculture, etc.), which directly impact the animal food chain through the consuming of contaminated water, plants and soil, and then other contaminated animals.

As detailed in the previous section, a wide range of chemical contaminants have already been classified with different physicochemical properties. They are scattered all around human and depending on its personal life style (food, inhabitation, work, etc.), external exposure varies from a subject to another. Major routes of external human exposure are commonly classified in three sections through i) ingestion (food, water, dust), ii) inhalation (indoor/outdoor air) and iii) dermal contact and application (soil, water, clothing, personal care products) (Elert et al., 2011). Figure 1.1 extracted from the US EPA EcoBox Tools by Exposure Pathways and Vermeulen et al., 2020 summarises this interlaced external exposure pathways constituting the human exposome through the environment-food-human continuum.

- 37/296 - Chapter 1. Non-targeted screening of biomarkers of human internal chemical exposure: state-of-the-art and challenges

Figure 1.1: Illustration of the environment-food-human continuum and external human exposure pathways. Adapted from US EPA EcoBox Tools by Exposure Pathways and Vermeulen et al., 2020.

The study of the external exposure allows to distinguish routes of exposure and to link exposure measurements with external determinants. However, this approach cannot establish links between exposure and health effect. Therefore, internal exposure assessment is required to evaluate the effect of the aggregation of routes of exposure. This approach is guided by human biomonitoring programmes and gives access to a global overview of population exposure. The present Ph.D. project falls within one of those programmes to better characterise the internal exposure in Europe.

1.1.4. Internal exposure and Absorption, Distribution, Metabolism and Excretion (ADME) processes

Following a given external exposure, contaminants penetrate in the human body and are then considered as contaminants of internal exposure. The internal exposure is characterised by three types of biomarkers named biomarkers of exposure, of effect and of susceptibility. According to the definition published by WHO/IPCS, 1993:

 biomarkers of exposure are exogenous substances, or its metabolites, measured in human body compartment.

- 38/296 - Chapter 1. Non-targeted screening of biomarkers of human internal chemical exposure: state-of-the-art and challenges

 biomarkers of effect are endogenous compounds secreted at different order of magnitude after an external exposure with possible health effect.  biomarkers of susceptibility are indicators produced as a response to external exposure depending of biological predisposition.

Once in the organism, chemicals can be absorbed and distributed according to their physicochemical properties and ability to cross biologic barriers. For instance, hydrophilic contaminants are mainly retrieved in blood and urine, whereas hydrophobic ones (e.g. POPs) are mainly stored in fatty tissues because of their lipophilic property (Lee et al., 2017) or retrieved in excretion compartments as human milk or faeces (Frederiksen et al., 2007). Some are rapidly eliminated as a native or metabolised molecule to facilitate the excretion (glucuronidation, sulfonation, hydroxylation etc.). Some others are stored in specific compartment (organs, tissues) for a certain period of time (days, months, years) and excreted according to the equilibrium with the concentration in bloodstream (Barr et al., 2005).

This complex interaction of contaminants with the organism leads to a very extended bioaccessible concentration range in the different biological matrices considered for exposure assessment. The identification of the most relevant marker/matrix pairs for biomonitoring and environment-health studies is a real challenge. The exhaustive way to assess the total internal exposure would be to analyse all body compartments but for obvious reasons it is not feasible to biopsy all organs. Consequently, biomonitoring study would be easier if biomarker of exposure/targeted matrix couples were already known thanks to a mapping approach but those studies are rare. A first largely studied matrix of interest is blood, as it is supposed to better reflect the global internal exposure susceptible to be further link to given health effects. Also, for practical reasons, it is easy to collect. However, it is invasive and requires medical assistance. On the contrary, urine may be directly collected by the patient and represents a part of excreted contaminants which can be correlated to the internal exposure and is accessible in larger volumes (Angerer et al., 2007). In order to extend this picture of internal exposure, complementary matrices (Sexton et al., 1995) may also be studied. For instance, adipose tissue (La Merrill et al., 2013), hair and nails (Papadopoulou et al., 2016) contain rich information regarding the past exposure, from weeks to years; placenta (Leino et al., 2013) and meconium (Ortega García et al., 2006) are characteristic of foetus exposure during pregnancy; human milk (Nickerson, 2006) illustrates the past mother exposure and one of the first baby contamination routes; other secretion fluids (saliva, sweat, tears (Esteban and Castaño, 2009), menstrual blood

- 39/296 - Chapter 1. Non-targeted screening of biomarkers of human internal chemical exposure: state-of-the-art and challenges

(Iribarne-Durán et al., 2020), seminal plasma (Saradha and Mathur, 2006), follicular fluid (Petro et al., 2012)) are means of body depuration. This non-exhaustive list of alternative biological matrices opens up the scope of possibilities for new research and discovery (Figure 1.2).

Figure 1.2: Generic view of contaminant absorption, distribution and excretion with different collectable matrices in blue. Adapted from Sexton et al., 1995 and Needham et al., 2005.

Beyond the fact of inter-individual variations in ADME process and genetic polymorphism, the daily exposure is also matter of high inter-individual variability. As an example, the flame retardant tetrabromobisphenol A (TBBPA) level in human milk varies from 0 to 688 ng/g lipid weight between studies in different counties over the world (Inthavong et al., 2017). This illustrates the complexity to characterise the human chemical exposome.

In addition to this global internal exposure, some sub-groups of population, experiencing pregnancy or childhood periods have raised significant concern. This was address during the present Ph.D. work by analysing biological human matrices such as placenta, human milk and meconium.

- 40/296 - Chapter 1. Non-targeted screening of biomarkers of human internal chemical exposure: state-of-the-art and challenges

1.1.5. Focus on early life stage of exposure

As a result of the previously described issues, individuals from the general and particular sub- populations are exposed to various complex patterns and levels of chemical exposure from conception to death. Many studies have reported health effect associated with this exposure such as increased risk of cancer, high cholesterol level and, immune system alteration or decrease of fertility, etc. (Fry and Power, 2017) and this impact of chemical environment on human health is a main current concern. Considering the extent of the health problem, the present research study was focused on early stage life of exposure including development periods. Many studies have already shown health risks due to exposure during pregnancy and/or childhood (Maccari et al., 2014). Exposure during perinatal period have been associated with hazard effects, for instance low term birth weight (Dadvand et al., 2013), and sometimes irreversible issues such as cognitive and motor development (Kyriklaki et al., 2016; Berghuis et al., 2018). This indicates that early exposure may impact the health at latter stages, including the health of the offspring or the ability to conceive. This trans-generational effect of chemicals is of high concern for pregnant and nursing women, where their own exposure is related to the foetus/newborn exposure via cord-blood and placenta transfer (Vizcaino et al., 2014) and breastfeeding (Fürst, 2006). On the other hand, breastfeeding has proven its importance for the newborn immunologic protection and human milk composition is adapted to the baby growth over the course of lactation (Oddy, 2001). Consequently, it is a priority to ensure good human milk quality for babies by controlling mother exposure.

1.2. Human biomonitoring and the HBM4EU project

1.2.1. Human biomonitoring programmes

The ongoing expansion of the exposome concept (Dennis et al., 2016; Jones, 2016, Niedzwiecki et al., 2019) and the enhanced development of related research activities, over the previous decade, reflect the increasing awareness of our environment as a source of human exposure to hundreds thousands of chemicals. This global contamination issue is a growing concern for exposure assessment programmes run by public health authorities (Louro et al., 2019).

In that context, scientists, chemical risk assessors and risk managers work in collaboration through human biomonitoring (HBM) programmes to assess the chemical exposure and the potential health effect in order to describe and then prevent and/or regulate the exposome

- 41/296 - Chapter 1. Non-targeted screening of biomarkers of human internal chemical exposure: state-of-the-art and challenges

(Figure 1.3). HBM programmes can be established at national or international level. For instance, USA has Centers for Disease Control and Prevention (CDC) initially opened in 1946 for malaria researches. In Europe, most countries have their own HBM programmes, for instance, Santé Publique France in France, Umweltbundesamt (UBA) in Germany, Flemish Center of Expertise on Environment and Health (VITO) in Belgium etc.

Figure 1.3: Addressing the Exposome through a trial collaboration between scientists, chemical risk assessors and risk managers over human biomonitoring programmes.

HBM programmes are useful tools (Bates et al., 2005) i) to evaluate time trends level of chemical exposure, ii) to highlight efficiency of regulatory decisions, iii) to map population chemical exposure over regions and iv) to characterise body burden distribution and establish reference values such as reference and exposure limit (German HBM I and II) values (Steckling et al., 2018). Consequently, those results support policymaking and action prioritisation. Analytical method reliability and assessment are thus crucial to assure quantitative results. Then, generated data should be comparable between sub-groups of population, analysed matrices and previously reported information. Hence, harmonised methodology at every stage is crucial too; from experiment design, to sample analysis and finally reporting of the results (Göen et al., 2012, Kolossa-Gehring et al., 2017). In addition, indicators understandable by non-expert audience can help to summarise all this scientific information, but are unfortunately lacking at the European Union level (Buekers et al., 2018).

- 42/296 - Chapter 1. Non-targeted screening of biomarkers of human internal chemical exposure: state-of-the-art and challenges

In 2004, the European Commission called for a joined HBM programme over Europe to harmonise scientific skills in chemical exposure characterisation and risk assessment. Thus, based on previous results from COPHES and DEMOCOPHES projects, the Human Biomonitoring for Europe project (HBM4EU) has been initiated (Ganzleben et al., 2017).

1.2.2. Human Biomonitoring for Europe (HBM4EU)

Global overview

The European project HBM4EU (www.hbm4eu.eu) was launched in 2017 for five years. This joint effort co-funded under Horizon 2020 gathers thirty countries, the European Environment Agency (EEA) and the European Commission to characterise the current exposome in Europe and the possible health effects in order to support policymaking. HBM4EU is coordinated by UBA and co-coordinated by VITO. Due to the large scale of the project, it is divided in three pillars, sub-divided in sixteen work-packages (WP) (Figure 1.4).

 Pillar 1 “Science to policy” acts as a bridge between science and policy to make sure that researches are adequacy with current policy questions and that delivered answers will help policy decision to protect human health.  Pillar 2 “European HBM platform” was created to fill the gap of harmonised information at European level. The final purpose is to centralise scientific skills in order to harmonise human biomonitoring activities through EU resulting in the establishment of a European platform of human biomonitoring.  Pillar 3 “Exposure and health” is generating HBM data to understand the exposure- health impact relationship, including CECs and mixtures of chemicals (Drakvik et al., 2020), in order to inform policy makers and public. The present Ph.D. thesis work is included within this last pillar, more precisely under the WP 16 “Emerging chemicals”.

- 43/296 - Chapter 1. Non-targeted screening of biomarkers of human internal chemical exposure: state-of-the-art and challenges

Figure 1.4: HBM4EU project skeleton.

Work-package 16 “emerging chemicals”

Nowadays, as previously mentioned, human is daily exposed to myriads of chemicals via various sources and routes such as environment, food or lifestyle, among which only few are a matter of dedicated quantitative analytical method, allowing the support and documentation of complete risk assessment processes. Health risks related to this exposure to chemical cocktails is another major societal concern. The vast challenge associated with the extent and diversity of this human chemical exposome requires a new generation of exposure assessment methods. Indeed, human biological monitoring has emerged during the last 15 years to provide precise internal exposure data for a number of known and a priori targeted chemicals. Although such data are still necessary, this traditional approach no longer appears compatible with the need of a wider exposure characterisation from very restricted biological specimens as typically available in human studies both for practical, analytical and economic reasons. In particular, the traditional targeted approaches are not suited to the detection of chemicals of emerging concern (CECs) for which large pieces of knowledge are still unavailable. Conversely, large scale suspect and non-targeted screening approaches based on current and future generations of

- 44/296 - Chapter 1. Non-targeted screening of biomarkers of human internal chemical exposure: state-of-the-art and challenges instrumentation dedicated to chemical profiling henceforth open the door to holistic characterisation of biological samples. This revolution enables the simultaneous detection of a never yet reached number of chemical descriptors, among which markers of human chemical exposure of interest for both exposure assessment, biomonitoring and environmental health studies. By encompassing those chemicals of emerging concern as well as unknown contaminants and their metabolites, these suspect and non-targeted approaches should provide a wider support to exposure assessment and early warning positioning.

The component of the HBM4EU project for which LABERCA is responsible (WP16) relates more particularly to the development and first proof-of-concept applications of such new methodological approaches for identification of chemicals of emerging concern. As described in the following sections, such large-scale suspect and non-targeted chemical profiling approaches based on chromatography coupled to high-resolution mass spectrometry (HRMS) are today available for holistic characterisation of biological samples. These advanced techniques allow the simultaneous detection of a large number of chemical features, including markers of human chemical exposure. At a regulatory and policy level, the challenge is to develop early warning capability to rapidly handle these chemicals through biomonitoring program and further risk assessment process. At a scientific level, the challenge first involves developing new methodological strategies to document the reality of exposure and the related health impact for these chemicals, then prioritising them on the basis of relevant and well- integrated exposure and toxicological data. This component of the HBM4EU project is a collaboration between seventeen partners over thirteen countries and is structured around different actions. The present Ph.D. thesis work is more precisely included within the action 3 “Non-targeted screening on storage/excretion “fatty” compartment” and action 4 “Suspect and non-targeted screening on foetal exposure related compartments”.

1.3. Definitions for setting the scene

Depending on the level of pre-existing knowledge associated with the considered biomarkers of exposure, three related methodological approaches can be used to stratify the human chemical exposome, namely i) targeted methods for known compounds, ii) suspect screening for known unknowns and iii) non-targeted screenings (NTS) for unknown unknowns (Figure 1.5). Suspects can be “converted” into targets by collecting comprehensive mass spectrometric reference data and retention indexes that enables unequivocal identification of the suspect

- 45/296 - Chapter 1. Non-targeted screening of biomarkers of human internal chemical exposure: state-of-the-art and challenges compound (usually reliant on the availability of reference standard compounds). The remaining signals in the sample are generally termed “non-targets” or “unknown unknowns”, for which no identity can be readily assigned, requiring further structural elucidation using non-targeted approaches.

Figure 1.5: Conceptual view of the human chemical exposome, related methodological approaches and associated purposes

1.3.1. Targeted screening

“Targets” are compounds of known chemical name and structure, for which quantitative targeted methods are available, alongside some exposure and risk assessment data (Smolders et al., 2009). Highly selective sample preparation is typically undertaken in order to isolate these targeted compounds with maximal removal of matrix interferences (Yusa et al., 2012). Detection and quantification is often conducted using low-resolution mass spectrometers (e.g. triple quadrupole - QqQ), usually operated in selected reaction monitoring (SRM) acquisition mode, to provide both high specificity and sensitivity (analytical background details are in section 1.4.1). Identification is supported by comparison with reference data acquired from certified standards (chromatographic retention time, MS and MS/MS spectra) used to validate compound identity prior to analysis. Quantification is preferably performed using the isotopic dilution method, enabling maximal performance with reduced uncertainty. A number of guidelines already exist to harmonise method performance assessment (e.g. 2002/657/EU for food). To some extent, targeted screenings can be also conducted using high resolution instrumentation (e.g. Orbitrap or time-of-flight), opening the door to simultaneous targeted

- 46/296 - Chapter 1. Non-targeted screening of biomarkers of human internal chemical exposure: state-of-the-art and challenges analyses with suspect and non-targeted ones (as described below) although some limitations may be encountered in this case compared to fully designed targeted methods based on tandem mass spectrometry (Cajka and Fiehn, 2016).

1.3.2. Suspect screening

“Suspects” are known compounds (“known unknowns”) in terms of chemical name and structure which are expected (“suspected”) to be present in a sample (e.g. occupational exposure). The typical approach applied in this case is large-scale suspect screening aiming to generate semi-quantitative data and contribute to better prioritisation for further targeted developments (Cortéjade et al., 2016) and confirm contaminant presence in sample. The same approaches are also helpful to elucidate the composition of complex mixtures by simultaneously generating exposure data for a wide range of markers of exposure from each individual sample. In most cases, analytical standards are not readily available and therefore, relevant analytical methods are not validated and compound identities not definitive. To some extent, suspect screening can be considered an extension of multi-class/multi-residue analysis, whereby some markers may be unambiguously identified and possibly quantified as per a targeted method, while others are mostly qualitatively measured. This qualitative annotation step refers to the assignment of a given compound identity to a signal detected by suspect or non-targeted approaches and relies on the elaboration and implementation of reference libraries to match the generated experimental data with structural descriptors indexed from a list of a priori defined chemical compounds.

1.3.3. Non-targeted screening

Non-targeted screening aims to detect “unknown unknowns” compounds without a priori criteria, to identify potential new markers of exposure and toxicological concern (Sobus et al., 2019). Generally, sample preparation and data acquisition are similar for suspect and non-targeted screenings whereas data analysis/mining are different. Although highly challenging due to paradigm shift from targeted analyses, this approach represents a very promising strategy to advance our knowledge of the human chemical exposome as it allows to extend the field of vision to unknown contaminants. In addition, it will enable better anticipation of future health threats and related risk assessment and regulatory dispositions. The development and implementation of NTS requires advanced capabilities and good integration of new front-of-science data management aspects (advanced data acquisition and processing

- 47/296 - Chapter 1. Non-targeted screening of biomarkers of human internal chemical exposure: state-of-the-art and challenges facilities, bioinformatics and modelling tools). A solid basic knowledge of chemistry (MS, NMR, chemical synthesis) and biochemistry is essential to allow the unambiguous identification and relevant interpretation and contextualisation of compounds besides the revealed signals. NTS is then coming with new paradigm modifying the conventional hypothesis-driven research approach to a data generating hypothesis-driven approach, as a really open way to characterise biological samples. It opens the detectable range of molecules contained in a sample and complete the capabilities of targeted approaches (Figure 1.6).

Figure 1.6: Chemical space detectable by applying a given chromatography coupled to mass spectrometry based on methodological screening workflow.

1.4. Challenges of non-targeted screening workflows

This section introduces the main challenges encountered through the development of a non-targeted methods from the sample preparation to the data processing. It has already been published in Pourchet et al., 2020 and is based on literature as well as observations and experiments conducted during the Ph.D. project.

1.4.1. Preliminary analytical background

Regarding either targeted or non-targeted approaches, the analytical workflow is constituted by three major steps, following pre-analytical phases such as sample collection and storage, namely 1) sample preparation, 2) analysis and 3) data processing.

 Sample preparation is based on extraction and purification to refine the sample in order to separate compounds of interest (i.e., markers of exposure) from interferences (e.g., lipids,

- 48/296 - Chapter 1. Non-targeted screening of biomarkers of human internal chemical exposure: state-of-the-art and challenges

proteins, other matrix components) and transform the sample into a final extract compatible with the analysis.

 Analysis is the detection of compounds of interest that involves different technologies. The most commonly used technique in the field of organic contaminant analysis at trace level in complex biological matrices is the chromatography coupled to mass spectrometry (MS). The principle is detailed in Figure 1.7.

Figure 1.7: Principle of chromatography coupled to mass spectrometry.

Sample extracts are introduced into a chromatographic system in order to discriminate the vast number of compounds present in complex biological matrices. They are separated according to the affinity with the stationary phase and either the mobile phase or the carrier gas, for liquid chromatography (LC) (Guillarme and Veuthey, 2017) or gas phase chromatography (GC) (Hübschmann, 2015) respectively. At the exit of the column, compounds are transferred into the mass spectrometer constituted of an ion source, an analyser and a detector. Once in the ion source, compounds are ionised at low or high energy, under different possible pressure. Most frequently, soft ion sources are electrospray (ESI) in LC and atmospheric pressure chemical ionisation (APCI) and atmospheric pressure photo ionisation (APPI) in both LC and GC, to generate molecular ion and adduct, whereas electronic impact (EI) in GC is an ionisation at high energy (commonly at 70 eV, but can be lower) generating many fragments. Then, ions are transferred into the mass analyser thank to lenses. The analyser is either selecting a targeted ion (single ion monitoring - SRM) or conducting a large group of ions (full-scan mode) according to their radio frequency. It also refocuses ions and amplifies the signal before transfer to the detector. In the detector, ions are recorded in function of the mass-to-charge (m/z) ratio. Most common (i.e. routinely applied) analyser-detector couples are single or triple quadrupole (Q

- 49/296 - Chapter 1. Non-targeted screening of biomarkers of human internal chemical exposure: state-of-the-art and challenges and QqQ) for low resolution mass spectrometer (LRMS), time-of-flight (ToF) and Orbitrap for high resolution mass spectrometer (HRMS). The Fourier transform ion cyclotron resonance (FT-ICR) is another powerful instrument type operating at ultra-high resolution but that remains the most expensive through the market and is not fully compatible with routine and high throughput analyses (Hoffmann and Stroobant, 2007).

The mass resolution is the detector capacity to distinguish two ions with very close m/z. The mass accuracy is calculated in part per million (ppm) as the ratio of the difference between the experimental mass and the theoretical mass out of the theoretical mass. ToF, Orbitrap and FT- ICR have high resolution up to 60 000 at m/z 200, 240 000 at m/z 200 and over 1 000 000 at m/z 400, respectively (Junot et al., 2014) with mass accuracy lower than 2 ppm for most recent ToF and lower than 1 ppm for Orbitrap and FT-ICR. Thus, HRMS devices allow to discriminate ions with close mass and therefore to detect more molecules with a good precision that reduce the number of theoretical chemical formula based on isotopic contribution.

 Data processing involves computational and bioinformatics tools to annotate, identify and sometime quantify compounds. Strongest identification by mass spectrometry is based on fragmentation (MS/MS) experiments. The ion of interest is selected in a first analyser, then fragmented with energy and some fragments of interest or all are selected in a second analyser (Figure 1.7). The resulting mass spectrum of fragmentation depends on the molecule chemistry and is characteristic for each compound. Depending on the energy applied to fragment the molecule the molecular ion is detected and some specific fragments too. This information is then implemented in a database for future identification. Annotation and identification are based on querying databases or library, that are lists of already known compounds containing different tiers of information (structure, fragment mass spectra, toxicity, etc.) and are generally developed for dedicated fields of interest (metabolomics, environment, forensic, etc.). Depending on the level of confidence, the compound is either annotated at lower level of confidence or identified when its identity is unequivocal.

1.4.2. Sample preparation

The selectivity versus sensitivity compromise

Suspect and non-targeted screening analytical workflows involve multiple steps ranging from sample preparation and data acquisition to data mining, expert reviewing and interpretation.

- 50/296 - Chapter 1. Non-targeted screening of biomarkers of human internal chemical exposure: state-of-the-art and challenges

Irrespective of the matrix of interest (i.e. conventional HBM matrices such as urine or blood, or alternative matrices including human milk, placenta, meconium or other tissues…), the first sample preparation step is critical and a compromise between selectivity and sensitivity has to be sought. In order to cover a wide range of potential markers of exposure with different physicochemical properties, extraction and purification should be as non-selective as possible. On the other hand, the sensitivity of the method is partly correlated to the concentration of matrix interferences which may impair the detectability and reproducibility of the signals of interest, for instance through ion suppression (Côté et al., 2009; Antignac et al., 2005). This also applies to ultra-HRMS instruments (e.g. FT-ICR) in direct introduction mode where in- source signal disruptions caused by matrix effects cannot be compensated by the high MS resolution of the detection system. Previous chromatographic separation can significantly reduce this issue, in addition to sample extract purification. Consequently, achieving an optimal level of compatibility of the prepared sample extracts with the instrumentation used for signal detection generally requires extraction step(s) followed by a certain level of purification. One of the main challenges associated with NTS is to achieve a balance regarding purification selectivity to limit matrix interferences (e.g. proteins, lipids, sugars, etc.), whilst preserving as many compounds of interest as possible with sufficient sensitivity. This new concept of cleaning to remove most abundant interferences is based on similar practical procedures used for targeted methods to enrich low concentrated analytes by purification but it is facing new issues, especially when the nature of the markers to be detected are not fully known in advance. This selectivity issue should remain a matter of flexibility to be adapted according to each specific application with regards to a particular interest toward certain chemical classes.

The starting sample volume compromise

The pre-analytical phases consist of sample collection and storage, which may also impact on the results obtained but which are not covered in the present section focused on the analytical steps. After the pre-analytical phase, a first crucial step that directly impacts the selectivity/sensitivity ratio is the selection of an appropriate starting sample volume for analysis. This volume depends on both the contamination level of the sample, which is partly unknown in NTS, and on the detection capability (sensitivity) of the instrumentation intended to be used. This parameter directly influences the possible enrichment factor of both markers of exposure (both known and unknown) and matrix interferences, as well as the efficiency required for the extraction and purification steps. As a general principle, the higher is the sample

- 51/296 - Chapter 1. Non-targeted screening of biomarkers of human internal chemical exposure: state-of-the-art and challenges volume considered for analysis, the more effective the purification step should be to concentrate markers of exposure and negate matrix interferences. This approach is commonly used for targeted analyses; especially for markers present at low concentration levels in complex biological matrices, for which relatively high quantity of sample is available (e.g. marine mammals, food, etc.). Conversely, human biomonitoring matrices are typically available in lower amounts than environmental or food matrices (µL/mL for human and mL/L for food and environment matrices), limiting possibilities for pre-concentration. Furthermore, the concentrations of environmental pollutants and/or their metabolites are typically orders of magnitude lower than concentrations of endogenous compounds, food constituents and drugs (Rappaport et al., 2014). Accordingly, as a general rule, NTS methods for human matrices should be preferentially based on low sample volumes and limited sample preparation. This paradigm shift from conventional targeted approaches leads to new analytical challenges, especially with regard to the sensitivity required for the MS detection systems, pushing toward high sensitivity HRMS instrumentation for accessibility to lowest concentrated exposure markers.

Extraction methods

Minimal sample preparation procedures, such as “dilute and shoot” or ultrafiltration, which are used in the metabolomics field (Khamis et al., 2017; Fernández-Peralbo and Luque de Castro, 2012), may also be applied to suspect and non-targeted screening of conventional biological fluids, such as urine or blood. These quick, simple and non-selective approaches have the advantage of preserving the sample integrity, limiting sample preparation related variability, and facilitating interlaboratory harmonisation. However, they are susceptible to significant matrix effects that can impair both the detectability of low concentrated markers of interest (Lin et al., 2010) and method repeatability. Online treatment is another option for limited sample preparation, for instance through turbulent flow chromatography (Couchman, 2012) or online solid phase extraction (Zhang et al., 2016).

Other human matrices, especially those with high protein and/or lipid content (human milk, faeces, adipose tissue, placenta, etc.) require a more elaborated treatment prior to analysis. For solid matrices, lyophilisation (freeze-drying) and/or grinding (e.g. via tissue lysers, freezer mills) allow sample homogeneity and extraction efficiency. Liquid-liquid and solid-liquid extraction (LLE/SLE) are the most commonly used approaches, where the nature and proportions of the applied solvent mixtures directly determine the range of extracted

- 52/296 - Chapter 1. Non-targeted screening of biomarkers of human internal chemical exposure: state-of-the-art and challenges compounds according to their physicochemical properties. In the context of suspect and non-targeted screenings, the selected extraction/partitioning solvents should allow the solubility of a wide range of compounds of differing polarity (Cajka and Fiehn, 2016).

In a first approach, a biphasic system can be suggested combining a polar solvent (water, methanol and/or acetonitrile) with one of intermediate polarity (e.g., diethyl ether, chloroform…), and/or one nonpolar (e.g. cyclohexane, pentane, toluene…). This partitioning allows collection of two or three complementary fractions from each analysed sample and is an efficient way to both divide (and so dilute) the whole matrix effect and enables characterisation of fractions by complementary technologies, e.g. LC-HRMS (predominantly hydrophilic compounds) and GC-HRMS (more hydrophobic compounds), respectively. The Bligh and Dyer approach originally developed for lipid extraction (Bligh and Dyer, 1959; Ulmer et al., 2018), consisting in applying a ternary solvent system with water, methanol and chloroform, is an example of such partitioning commonly applied and/or adapted for suspect and non-targeted screenings. Other closely related approaches are the Folch method (Folch et al., 1957) or more recently methanol/methyl tert-butyl ether (MTBE) method (Matyash et al., 2008). As a general rule, such non-selective procedures may be privileged to preserve sample integrity with maximal potential for exposure marker detection without a priori. Another approach consists in the Quick, Easy, Cheap, Effective, Rugged, Safe extraction (QuEChERS), which is well established for contaminant multi-residues analysis (Cloutier et al., 2017) and metabolomics (Garwolińska et al., 2019).

Liquid-based extraction may also be accelerated by using high temperature and/or pressure devices such as pressurised liquid extraction (PLE). This approach is adequate for (semi-)solid matrices and has proven its efficiency for targeted measurement of lipophilic contaminants (dioxins, PCBs, BFRs…) from environmental and food matrices (Vazquez-Roig and Pico, 2015). Microwave assisted extraction (MAE) is another alternative compatible with either solid or liquid matrices (Llompart et al., 2019).

Introduction of external contamination background not previously visible through the targeted approaches, or the possible degradation of some exposure markers of interest under these extraction conditions, are additional concerns for suspect and non-targeted screenings. These techniques may induce some selectivity with regard to certain classes of compounds, but also higher efficiency of the extraction process, which can be relevant in case of more oriented researches or applications. Finally, a rigorous assessment of the applied protocol appears

- 53/296 - Chapter 1. Non-targeted screening of biomarkers of human internal chemical exposure: state-of-the-art and challenges necessary and should be promoted in order to systematically document the application scope of the applied methods, i.e. evaluate and communicate on its suitability to detect only a certain range of compounds.

Additional purification and fractionation

To achieve better sensitivity and detectability of some markers of exposure in complex matrices, an additional purification may be required following the extraction step. In the present case of research of markers of exposure, lipids and proteins are considered as major matrix interferences. Introducing selectivity towards particular classes of chemicals can be justified by the specific research priorities of the developing laboratories. This may also be a pragmatic choice to maximise the sensitivity compared to generic preparation. However, the introduction of supplementary sample preparation steps may impair the global ambition of a non-selective, large coverage analysis due to the loss of some exposure markers of possible interest. Additional purification strategies may also compromise the fast and high throughput objectives expected for these NTS approaches.

Solid Phase Extraction (SPE) or d-SPE (dispersive-SPE, Bakhytkyzy et al., 2020) is an example of purification techniques widely used for targeted analyses that can be also applied for NTS (Samanipour et al., 2018). The selectivity of the resulting purification may be adapted according to the nature of the stationary and mobile phases, with a very large number of options today available. This approach is selective of compounds of interest which is not fully compatible with NTS. In this context, purification should be selective of matrix interferences instead of compounds of interest. Historical options are available such as organic solvent-based protein precipitation but also more recent options are developed to selectively remove particular matrix components such as lipids (e.g. Captiva ND / EMR SPE) (Zhao et al., 2018). Other approaches used in omics fields, such as chemoselective probes to target chemical groups (e.g. halogens) (Mitchell et al., 2014) or deproteination by applying magnetic beads (König et al., 2013) will maybe more developed in the future and could be adapted to NTS. In the same idea of a selective purification, sample fractionation can also be used for NTS, as a conservative clean-up strategy and/or for the confirmation of chemical structure besides the detected markers. Size Exclusion Chromatography (SEC) may be employed to this end (Saito et al., 2004). Sample fractionation is particularly applied when the chemical NTS is coupled to effect directed analyses (EDA). It aims to characterise a biological activity associated with the corresponding fractions in order to

- 54/296 - Chapter 1. Non-targeted screening of biomarkers of human internal chemical exposure: state-of-the-art and challenges highlight compounds responsible for the biological activity and to prioritise their identification (Simon et al., 2015)

Extract reconstitution

After extraction, the extract is usually concentrated by solvent evaporation (mostly to dryness, or by introducing a keeper solvent to preserve the loss of some volatile compounds). Subsequently, the last step before injection into the analytical instrument used for detection is reconstitution of the final extract. Technical consideration of reconstitution is not trivial and appropriate solvent(s) selection should be based upon the capability to re-suspend the extracted compounds and compatibility with the separation and detection system. The selected solvent (mixture) should be suitable to correctly dissolve non-targeted compounds and therefore to introduce them in the analytical instrument for their detection. Because suspect and non-targeted screenings aim to cover compounds with a wide range of physicochemical properties, the use of a mixture of solvents with complementary polarities and solubilisation capabilities is an appropriate strategy to reconstitute the final sample extract. In addition, the reconstitution can be ultrasonically assisted to reduce the adsorption of compounds on glass vial. The solvent (mixture) also needs to be compatible with the chromatographic system used for MS analysis, as the injection solvent system may greatly influence retention times, as well as peak shapes (e.g. water, methanol, acetonitrile, etc. in LC and toluene, hexane, etc. in GC). If no definitive guideline can be proposed at this stage, the systematic evaluation of the used reconstitution conditions on a set of QA/QC reference compounds covering the range of markers susceptible to be addressed by the method is a good option.

1.4.3. Analysis

Several options exist to analyse CEC by chromatography coupled to HRMS. LC-HRMS is at the present time the most commonly used technology for suspect and non-targeted screenings, either with Orbitrap or ToF devices. The present section highlights the complementarity of LC- HRMS and GC-HRMS (Figure 1.8) for covering a wide range of compounds in terms of molecular size, polarity and volatility (Pico et al. 2020). However, LC-HRMS remains the most used instrumentation for non-targeted screenings and is more detailed below.

- 55/296 - Chapter 1. Non-targeted screening of biomarkers of human internal chemical exposure: state-of-the-art and challenges

Figure 1.8: Complementarity of LC-MS and GC-MS based systems in terms of detectability of compounds with various volatility and polarity. Adapted from Brack et al 2016.

Chromatographic separation

For LC separation, a large diversity of stationary phases, mobile phases and solvent additives are available. Despite this diversity, reversed-phase columns (mainly C18) remain the most commonly used due to their efficiency over a wide hydrophobicity range. Their widespread use also allows methodological comparisons and harmonisation.

However, reversed phase chromatography with embedded polar groups or hydrophilic interaction chromatography (HILIC) are increasingly emerging alternatives for highly hydrophilic compounds, due to their orthogonality to the classical reversed phase chromatography (C18) and their compatibility with common ionisation sources (Jandera and Janas, 2017). Regarding the mobile phase composition, conventional water/methanol or water/acetonitrile binary systems are most commonly used, and both seem suitable (Yusa et al., 2015). A ternary system water/methanol/acetonitrile may also be suggested to take simultaneous benefit of the respective properties of both organic solvents. Because of the typically applied limited sample preparation, the use of a generic elution gradient can be recommended as a general rule to reach a satisfying separation of the analytes and to limit matrix effects. The introduction of a final flush of the column (e.g. with acetone/isopropanol) can also be advised to avoid carry-over between injections. Modifiers (acetic acid, formic acid, ammonium acetate, ammonium fluoride, etc.) are often added to the mobile phase in order to stabilise the pH, improve peak shape or promote ionisation or a specific adduct formation (Kruve and Kaupmees, 2017). The nature of the solvent modifier directly impacts the obtained

- 56/296 - Chapter 1. Non-targeted screening of biomarkers of human internal chemical exposure: state-of-the-art and challenges chemical profiles: the distribution of the different ionic species formed for the expected exposure markers (e.g. (de)protonated molecules vs. formate and ammonium adducts) makes the annotation process of the detected signals more complex and has to be handled by the data processing component of the analytical workflow.

Supercritical fluid chromatography (SFC) also constitutes another interesting tool, not yet routinely implemented in laboratories due to more recent re-interest, to separate a broad range of polar molecules (-2 < log P > +2) and may be considered as a “green” analytical technique (Losacco et al., 2019).

For GC separation, choices in stationary phase and carrier gas are less abundant than for LC. The nonpolar capillary column with 5% phenyl methylpolysiloxane is largely applied for discriminating environmental contaminants. Regarding carrier gas, helium, hydrogen and then nitrogen are widely used. Importantly, helium is preferable for safety reasons compared to hydrogen, which is flammable and for efficiency criteria compared to nitrogen, which offers slower separation. However, helium is a non-renewable resource on earth and its price increases.

Unlike LC where the solubilised extract is injected directly into the column, in GC, the extract is vaporised prior to deposition. This can lead to premature clogging of both injector and column, if the extract contains low or non-volatile compounds. As opposed to LC, it is not feasible to clean the column after each injection. However, other maintenance precautions are possible as an isotherm at high temperature and by coupling a deactivated fused silica tubing to the front of the analytical column. It allows to analysed less purified samples and to protect the analytical column from contamination. Two approaches are generally used, either coupling 10 m of guard column and cut 1 m after each injection batch, or coupling 2 m of guard column and systematically replace it after each injection batch. The first option appears more convenient as manual interventions are required. Indeed, coupling a guard column increases the risk of leak and this risk more present if the guard column is systematically replaced. Using a 10 m of guard column and cutting it affects the retention time (rT) as the length of column is shorter. The rT is an important piece of information with regard to the identification of the corresponding markers. Although the marker’s identification is more often primarily based on spectrometric characteristics (exact mass, isotope and fragmentation patterns), the experimentally observed vs reference or predicted/modelled rT is helpful to decrease the number of candidate chemical structures possibly fitting with a given detected accurate mass. Thus, coupling a guard column

- 57/296 - Chapter 1. Non-targeted screening of biomarkers of human internal chemical exposure: state-of-the-art and challenges should be used with a rT correction to make data comparable such as the introduction of retention index (RI) (Celma et al., 2018). This mathematical approach was first introduced by Kovats in 1958 for isothermal gradient and updated in 1963 by van Den Dool and Kratz for linear gradient. It allows to correct the rT with a universal standard mix commonly acid methyl esters (FAMEs) or alkanes mix (from C8 up to C40) by calculating a value depending of the compounds rT in comparison with standards elution. This approach is already well- implemented for GC-MS profiling and most of compounds are recorded in libraries, such as NIST, with this retention information.

Whereas rT takes a great part of the compound identification in GC-MS, improvements are required for LC-MS. As the retention time is influenced by the stationary phase, mobile phase, temperature, flow rate and column dimensions, the harmonisation of rT in LC does not appear realistic over all developed methods. LC retention projection (Abate-Pella et al., 2015) was proposed to calculate LC elution gradient retention across laboratories based on the isocratic retention factor versus solvent composition relationship already recorded in some database. This approach appears promising to increase the confidence level of identification and should be considered for database development. Furthermore, as the measurement the isocratic retention factor versus solvent composition relationship is required, the development of an easy and fast calculation is expected.

Ionisation and detection

First of all, electrospray ionisation (ESI) is the most widespread ion source for LC. The present manuscript will thus focus on this technological option, although alternatives such as APCI and APPI may have some advantages in terms of marker’s coverage. Using ESI with both positive and negative ionisation modes would maximise the number of detected markers, either through two separate injections or using a single injection in the polarity switching mode for MS devices with sufficiently high scan rates.

Then, high-resolution mass analyser/detector is required for NTS to reach unique elemental composition in order to facilitate data processing and improve marker of exposure identification. LC-HRMS coupling usually refers either to Orbitrap or to time-of-flight (ToF) HRMS devices. Potentially, Fourier-transform ion cyclotron resonance (FT-ICR) instruments may be applied, especially with regard to their structural elucidation capabilities based on ultra- high resolution (Kind and Fiehn, 2006). However, FT-ICR may face some limitations with

- 58/296 - Chapter 1. Non-targeted screening of biomarkers of human internal chemical exposure: state-of-the-art and challenges regard to in-source matrix effects and related signal non reproducibility that impair their application for complex matrices where signals of interest are of very low abundance, as it is the case for HBM. Their elevated cost also limits their large-scale implementation, and so this option cannot be really considered in the current state as a priority for laboratories aiming to implement NTS, nor suitable for short-term method harmonisation.

In terms of mass ranges, priority can be given to the m/z 50-1000 range, fitting with the properties of exposure markers typically expected in HBM. Indeed, most of chemical contaminants (e.g. compounds listed in the Stockholm convention) are included in the mass window. However, the m/z 1000-2000 range may also be informative and optionally covered to detect additional markers, as well as contribute to confirm marker identity through the detection of supplementary adducts for high molecular weight compounds.

Besides the ionisation polarity and mass range criteria, the full-scan mode acquisition represents the starting recommendation for NTS. At this stage, ensuring maximal reliability of the generated data appears crucial especially in terms of resolution and mass accuracy. Appropriate control and adjustment for these two settings are necessary, through suitable calibration procedures, as well as through the recurrent analysis of appropriate mixtures of reference compounds/ material. For state-of-the-art instrumentation, mass resolution typically exceeds 30,000 and mass accuracy is below 5 ppm. More advanced/latest generation of instrumentation reach rather higher performance and should be preferred for NTS in order to facilitate data processing.

Data dependent acquisition (DDA) and data independent acquisition (DIA) are more advanced options that have to be considered for generating structural information in the context of NTS (Oberacher and Arnhard, 2015). These acquisition modes require hybrid MS instruments equipped with fragmentation capabilities. Briefly, DDA is more restrictive than DIA, by fragmenting the “n” only ions passing a specific threshold (e.g. abundance, neutral loss etc.) of the MS full-scan. In contrast, all ions are fragmented in DIA and dedicated data treatment tools are required (e.g. SWATH developed by Sciex or HRM by Biognosis (Ludwig et al., 2018)) to properly assign the various fragment ions to their respective precursors. This increases the complexity of deconvolution in DIA and becomes a major limitation in the data processing step. The additional information related to the structure of the compound, in addition to its exact mass (elemental composition, isotopic pattern), is the basis of an increased confidence level for

- 59/296 - Chapter 1. Non-targeted screening of biomarkers of human internal chemical exposure: state-of-the-art and challenges compound identification. Nevertheless, the advanced data acquisition is currently still a matter of research and development, and depends on the considered generation of instrumentation.

Finally, developing methods based on non-selective data acquisition modes is in agreement with the objectives for suspect and non-targeted screenings. As for the development of the sample preparation method, the analytical approach appears as a paradigm shift from conventional analytical approaches, leading to new challenges (Figure 1.9). For suspect and non-targeted screenings, no global harmonisation related to the data acquisition can be recommended. Conversely, complementarity and orthogonality of various analytical methods present obvious advantages to identify as many markers as possible. The effort should preferably focus on the development of annotation MS reference libraries with sufficient flexibility to cover the different ionic species potentially expected for the different markers under various analysis conditions.

- 60/296 - Chapter 1. Non-targeted screening of biomarkers of human internal chemical exposure: state-of-the-art and challenges

Figure 1.9: Summarised conceptual comparison of the analytical workflows typically applied for conventional targeted (left) versus non-targeted (right) methods (A), and of the resulting global performance expected with both approaches (B). Radar chart axis scale is in arbitrary units with 0 correspond to low performance and 100 a high performance.

- 61/296 - Chapter 1. Non-targeted screening of biomarkers of human internal chemical exposure: state-of-the-art and challenges

Effect-directed analysis

In addition to mass spectrometry detection, biological test activity offers complementary information about contaminants able to link with biological receptor. Effect-directed analysis (EDA) is a biological key approach to screen markers of exposure regarding their direct effect with organism (Brack et al., 2016). Following the sample preparation, the extract is injected on a LC system for a chromatographic separation and is either detected in MS or transferred to a collector device. In the latter, the extract eluted from the LC column is fractionated in a 96-well plate and because it is the same chromatographic separation, each well corresponds to a retention time window in LC-MS. Then, the biological activity (e.g. thyroid hormone system disruption) can be measured by fluorescence in each well and the response can be interpreted thanks to the corresponding mass spectra (Figure 1.10) (Booij et al., 2014). This approach is of interest for non-targeted screenings, because it focuses on a sub-group of chemicals according to their health impact without a priori knowledge and it is able to screen new markers of exposure interacting with biological mechanism, such as hormonal system.

Figure 1.10: Principle of effect-directed analysis combining chemical and biological approach. After the sample treatment, extract is separated by liquid chromatography and the post- column flow is split in two. One part is dedicated to HRMS analysis in full-scan mode and the other part is fractionated in a 96 wells plate. The biological activity of each well is measured with the biological test of interest and the biological response is interpreted with regards to MS data.

In practice and for the present case, the thyroxine hormone (T4), a thyroid hormone, is transported and distributed through the human body bound to transport proteins like transthyretin (TTR), among others. According to the molecular geometry, other compounds such as PCBs, PBDEs, Br/Cl-phenol, can interact with the TTR and create a binding competition. These competitors to T4, and other thyroid hormones such as triiodothyronine (T3), are classified as thyroid hormone system disruptors. In order to measure this competition

- 62/296 - Chapter 1. Non-targeted screening of biomarkers of human internal chemical exposure: state-of-the-art and challenges and thus detect thyroid disruptors, the TTR-T4-FITC bioassay activity is used. A luminescent probe fluorescein isothiocyanate (FITC) is bound to T4 hormone and when this complex is conjugated to TTR protein, the luminescent probe emits a more intense fluorescence response than when it is not conjugated to the TTR protein (Ren and Guo, 2012). To assess thyroid disruptor presence, the final extract resulting from sample preparation is in contact with TTR protein and T4–FITC complex. If thyroid disruptors are in the extract, they enter in competition with the T4-FITC complex to be bound to the TTR protein and the fluorescence response decreases (Figure 1.11).

Figure 1.11: Fluorescence response of the T4-FITC complex (red triangle) bound to TTR protein (blue hexagon) as a function of the thyroid hormone system disruptors (yellow hexagon) concentration.

This EDA approach is in particular well-known in the Environment and Health department of Vrije Universiteit Amsterdam (VU-E&H). In the frame of the HBM4EU a collaboration between WP16 partners (LABERCA and VU-E&H) was conducted through Ph.D. student secondment, including my own stay during three weeks at VU-E&H under the supervision of Pf. Marja Lamoree. This collaboration was focused on EDA approach with bioassay specific of EDC and more precisely of thyroid hormone system disruptors. EDC are compounds of concern because of their effect on health for instance impact on male and female reproductive health, thyroid-related disorder, neurodevelopmental disorders in children. Thyroid hormones have important role in the development, including during the perinatal period. Thus, it appears important to monitor those thyroid hormone system disruptors through biomonitoring

- 63/296 - Chapter 1. Non-targeted screening of biomarkers of human internal chemical exposure: state-of-the-art and challenges programmes with dedicated approach. This EDA approach has already demonstrated its efficiency on blood of polar bear to identify thyroid hormone system disruptors (nonylphenol, mono- and di-hydroxylated octachlorinated biphenyls) (Simon et al., 2013).

As previously explained, a single compound could have a moderate health effect when considered alone but becomes hazardous in combination with other chemicals. This mixture effect is not accessible when the sample is fractionated. However, when this bioassay is conducted directly after the sample preparation on the whole extract, it offers information about mixture effect and is interesting from a toxicological point of view. Furthermore, used as a first screening, the whole extract bioassay enables to prioritise samples regarding their level of contamination. In a context of human biomonitoring and in order to screen new markers of exposure, it would be interesting to first analyse the whole extract and then focus with the fractionation approach on highly contaminated samples.

1.4.4. Data processing

LC- and GC-HRMS data acquired in full-scan mode are constituted by a vast number of detected signals. Data processing is therefore a necessary, crucial and long step in order to extract relevant portion from this wide volume of information, in accordance with the research question of interest.

Post-acquisition data pre-processing

In the context of suspect and non-targeted screenings, the post-acquisition data pre-processing step consist in shifting from raw instrumental data to a tabulated curation file (peak list containing at least, rT, m/z, intensity values) used for subsequent annotation and statistical analyses. This appears as the main component of non-targeted analyses and represents a substantial effort, since it can be very labour intensive and time-consuming, often requiring manual intervention/oversight. This component crucially depends on the availability and performance of bioinformatics tools. A wide range of software is available to perform the extraction of information from the raw data (Stanstrup et al., 2019; Hu et al., 2016). Some of these tools are integrated solutions from MS vendors, e.g. Metaboscape from Bruker, Progenesis QI from Waters, Trace finder, Sieve and Compound Discoverer from Thermo, Mass Profiler Professional from Agilent, MetID from Shimadzu, or XCMSplus from Sciex. Other options are open source software (Spicer et al., 2017), many largely implemented in the

- 64/296 - Chapter 1. Non-targeted screening of biomarkers of human internal chemical exposure: state-of-the-art and challenges metabolomics community, e.g. XCMS in the R computational environment / online (Tautenhahn et al., 2012), MZmine 2 (Pluskal et al., 2010), Workflow4Metabolomics (Giacomoni et al., 2014), MS-DIAL (Tsugawa et al., 2015) and MetAlign (Lommen and Kools., 2012). Other in-house developed solutions complete this panel of existing offers, such as HaloSeeker (Léon et al., 2019), an open access tool designed for the specific screening of halogenated markers.

These data processing tools aim to detect any signal present in the generated chemical profiles (peak picking), to align common peaks found in the different samples (peak alignment) and report their intensity or area (peak integration). In practice, the settings for the respective algorithms have to be carefully chosen, as they directly impact the obtained information and even induce some pitfalls on the generated results. Even after years of use and experience with some of these tools, there are still no consensus guidelines regarding both a preferable selection from this panel or their fine appropriate parameterisation in the context of NTS. As settings also depend on the analytical configuration (LC/GC and MS settings), a harmonised procedure is also hard to implement. This absence of comprehensive and universal data processing solution still appears as a main bottleneck of NTS approaches (Baran, 2017; Tugizimana et al., 2016). For HBM, the limited possibility of sample replication, commonly applied in metabolomics to manage variability, complicates this component of the NTS workflows. Therefore, the establishment of common QA/QC measures to reach a better level of confidence on the produced results and a better comparability between different data processing approaches appears to be necessary (Considine et al., 2018). Defining and reaching correct data processing outputs for a set of QA/QC reference compounds covering the range of markers susceptible to be addressed by the method here appears as a good option.

Compound annotation and identification

Importantly, the confidence level associated with the identification of the detected markers depends on the type and extent of structural information collected and available through the implemented analytical workflow. A harmonisation proposal was elaborated in the water analysis and metabolomics communities to clearly distinguish the levels of confidence (Figure 1.12), from level 5 where only exact mass is available to describe the considered marker, to level 1 where full mass spectrometric pattern (MS/MS data) and rT are available and successfully compared to an analytical standard (Schymanski et al., 2014). Intermediate confidence levels are reached through the querying of databases where chromatographic and

- 65/296 - Chapter 1. Non-targeted screening of biomarkers of human internal chemical exposure: state-of-the-art and challenges

MS descriptors (so-called “MS ready information”) are indexed for a list of a priori defined chemical compounds (McEachran et al., 2018; Oberacher et al., 2019; Schymanski and Williams, 2017).

Figure 1.12: Identification confidence level proposed by Schymanski et al., 2014 for HRMS data.

More recently, the Metabolomics Society proposed another standard to report metabolite annotation including information about the stereochemistry and chirality of the molecule (reference and illustration are in Appendix 1). The Metabolite Identification Task Group was specifically initiated to cover this topic and build a consensus with the community. However, this new scale is not adopted yet and thus not widely applied. This demonstrates the difficulty to establish novel standards and the need to continuously adapt the previously existing guidelines to new topics of research.

Spectral properties can be either experimentally determined from analytical standards, or theoretically calculated/modelled through bioinformatics tools (e.g. MS/MS spectral similarity networks oft termed as “molecular networking” or fragment tree correlations, network propagations via e.g. substructure/motif searching or in silico fragmentation). The same marker’s ID confidence scale should be then more largely adopted and harmonised in the exposomics community. Such databases are well developed in the metabolomics community, primarily focused on endogenous compounds acting as markers of effect (e.g. Human Metabolome DataBase (HMDB), and METLIN (Warth et al., 2017)). Although not directly suitable for annotating markers of chemical exposure, these metabolomics-related databases

- 66/296 - Chapter 1. Non-targeted screening of biomarkers of human internal chemical exposure: state-of-the-art and challenges can be useful in the annotation pipeline to reveal and then discard from further processing, endogenous compounds (metabolites of endogenous substances, markers of effect…). The US- EPA Comptox Dashboard is another existing resource in the field (McEachran et al., 2017) incorporating an environmental compounds component, as well as some particular sub- databases focusing on particular classes of substances, for instance psychoactive substances (Mardal et al., 2019; Lung et al., 2016). Yet, no extended and consolidated MS reference library exists at European scale to annotate markers of chemical exposure (either parent compounds and/or their metabolites) and to accompany the development of the exposomics field. One ambition of the component of the HBM4EU initiative dealing with these new methodological approaches is to build this ambitious and QA/QC consolidated database dedicated to markers of human internal exposure to CECs (see details in section 2.4.1.4. HBM4EU database). It also aims to develop a data processing methodology to prioritise the way to analyse the generated data (peak-picking, pairing, alignment, background subtraction etc.), in the spirit of NTS based on non a priori assumptions.

Finally, the confidence level associated with the markers identified through suspect and non- targeted screening approaches may be highly variable from one study to the other, as well as from one given marker to the other within the same study. Until the availability of more consensual standards, a careful documentation of the real level of identification associated with each reported marker appears mandatory in this emerging field. Harmonised reporting of suspect and non-targeted screenings results also appears as a priority and a way to clarify and make transparent and comparable this crucial issue of marker’s identification from a given study to another one, especially in regulatory and support to policy contexts. Development of a common reporting template for European exposure is an ongoing activity developed within the HBM4EU initiative to report suspect and non-targeted screening results, as it was already done by EPA with ENTACT initiative (Ulrich et al., 2019)

1.4.5. Method performance assessment

The analytical chemistry community was proficient for many years in the assessment of the performance of conventional targeted methods with appropriate QA/QC measures (Figure 1.13). One of the main approaches uses one or more reference standard compounds to evaluate (then possibly validate) various analytical criteria including selectivity, recovery, accuracy, linearity, limits of detection and quantification, etc. Those criteria are assessed for known and

- 67/296 - Chapter 1. Non-targeted screening of biomarkers of human internal chemical exposure: state-of-the-art and challenges targeted molecules so that the method is validated for a set of compounds. Consequently, non-targeted method assessment is facing a more complex situation, since some of the signals of interest are still unknown. However, several concepts from targeted analysies can be transferred to the non-targeted methods. Therefore, non-targeted workflows will be fit for purpose if they are able to reliably confirm the presence of predefined chemicals being representative for the chemical domain of interest in defined biological materials at concentration levels typically observed in humans exposed to those chemicals (Sobus et al., 2019). An appropriate set of such QA/QC samples should include system suitability test samples, fortified and/or naturally contaminated matrix samples, as well as procedural blank samples. This QA/QC aspect is well implemented in the last generation of metabolomics studies (Dudzik et al., 2018), and consequently, it should also be better developed in the exposomics area.

Figure 1.13: Summarised conceptual comparison of the QA/QC current state of development typically observed for conventional targeted (left) versus non-targeted (right) methods.

Fortified samples with a set of known substances are useful for various QA/QC purposes. During method validation, they are used to test detection capabilities, reproducibility, as well as reliability of identification. Furthermore, spiked samples are used as QC samples to monitor performance over a batch of samples and for batch-to-batch corrections (i.e. stability of retention times, chromatographic performance and peak shapes, mass accuracy and resolution, detection sensitivity, stability of signal intensities). A set of known markers of exposure covering a broad range of physicochemical properties, selected to be representative for the expected diversity of marker compounds can be used as indicators of the method performance

- 68/296 - Chapter 1. Non-targeted screening of biomarkers of human internal chemical exposure: state-of-the-art and challenges at various levels, covering sample preparation (recovery, matrix effect…), data acquisition (chromatographic and mass spectrometric resolution, mass accuracy…) and data processing (peak picking and alignment…) steps.

Another QA/QC related issue associated with NTS approaches is the assessment, control, and management of the external contamination encountered in the procedural blank samples. The issue proves to be more problematic for NTS than for conventional targeted approaches because non-selective sample preparation and data acquisition can lead to detection of various compounds originating from sources other than the sample itself (e.g. plasticisers, plastic additives, solvent/reagent impurities…). It is important to define which part of the generated information relates to the analysed sample and which to instrument noise or external contamination. In practice, there are a number of difficulties related to the characterisation of the background noise and many ways to manage it with a well-established, reliable, and documented blank subtraction process (Caesar et al., 2018). Several parameters need to be established, such as the number of procedural blanks introduced in each analytical batch and the method to assess them. Another crucial question is how to establish a reliable limit of reporting for compounds present in the blank and in higher concentration in the sample? All these points require significant efforts in targeted methods and are clearly also relevant for NTS. Even in established workflows, these issues are not always adequately considered (Considine et al., 2018; Dudzik et al., 2018; Boccard et al., 2010). It requires not only strict analytical precautions, but also new conceptual and computational solutions with regard to data handling, normalisation, statistical treatment etc., and will require additional collaborative work in order to achieve better harmonisation. Meanwhile, appropriate documentation of the procedures for characterising and manage the external procedural contamination in NTS is needed.

Evaluation of NTS approaches can also be performed through comparison to results obtained by targeted approaches. This is recommended to be implemented during method development and represents a valuable way to consolidate the NTS workflow, by accompanying any new analytical option tested with a given “reference” result. Even if conducted for a limited number of markers, this approach is useful to better qualify the NTS performance, including in particular for a first evaluation of the false negative and false positive rates (McCord and Strynar, 2019; Herrera-Lopez et al., 2014). Coupling of traditional and emerging methodologies will allow an efficient mutual benefit for both biomonitoring and exposomics areas (Dennis et al., 2016).

- 69/296 - Chapter 1. Non-targeted screening of biomarkers of human internal chemical exposure: state-of-the-art and challenges

1.5. Conclusion

The present chapter illustrates the wide expanse of the chemical exposome area and the difficulties encountered for its characterisation. External exposure assessment is complex because of numerous interlaced pathways of exposure occurring at different life stage and depending on individual life style. Internal exposure is the reflection of multiple external exposures over hours and even years. Studying the internal exposure is also challenging regarding the panel of different human matrices available and their intrinsic properties to metabolise and/or store and/or eliminate contaminants. After decades of research to investigate the chemical exposome, many methods have been developed under human biomonitoring programmes to address such issues. The golden approach consists in targeted methods to quantify known contaminants. However, complementary approaches able to screen a broader range of contaminants are required in order to characterise the evolution of the exposome constitution. Non-targeted approaches based on large screening of compounds without a priori knowledge are emerging in human biomonitoring projects. The development of such approaches is characterised by both a contextual scientific background of high complexity and an underlying necessary methodological framework of high technicity. There are significant challenges for this field that entail major analytical developments related to each step of the workflow. The analysis of human samples requires specific methodologies and processes compared to other fields of application, such as environment or food. On one hand, rigorous harmonisation measures are required to achieve better consolidation and comparability of data generated from various studies, especially regarding their further use in a regulatory and support to policy context. On the other hand, this evolving field requires considerable flexibility in order to maintain its capacity in discovery and exploratory research.

Despite the lack of harmonised procedure to develop NTS method, previously described challenges offer broad guidelines:

 Besides particular cases, sample preparation should provide minimal selectivity to encompass the desired diversity of exposure markers and an acceptable purification for limiting matrix interferences and their detrimental impact on the overall method performance.

- 70/296 - Chapter 1. Non-targeted screening of biomarkers of human internal chemical exposure: state-of-the-art and challenges

 Considering the wide range of relevant markers of exposure in a whole NTS context, the combination of several complementary sample preparation methods is beneficial and global harmonisation on that point does not appear a priority.  The implementation of these approaches requires high-level equipment and significant technical expertise, with an optimal combination of both LC-HRMS and GC-HRMS to achieve broad coverage of markers with various physicochemical properties.  Considering the large number of possible technical choices and parameterisation options, no strict guideline with regard to HRMS data acquisition for NTS can be elaborated, yet some harmonisation is desirable and appears to be possible.  Data processing applied to NTS requires advanced computational tools many still under development and represents one of the major challenges due to the highly complex data, thus necessitating specific expertise and a critical view of the generated results to ensure consolidated outputs. There is a critical need in the field for multidisciplinary as well as for high level and sustainable competences, which are not traditionally present in analytical laboratories.  The development of an extended and qualitatively consolidated MS reference library for annotating markers of exposure is a key strategical element for operational and harmonised implementation of these approaches at the European scale.

In that context, the present Ph.D. thesis project addressed each point by developing an analytical method from the sample preparation, analysis to data processing applied to the characterisation of human matrices. It aims to lay the foundation stones of non-targeted approaches to support human biomonitoring programme and to highlight relevant and promising perspectives. In front of this ambitious project, tasks were prioritised to characterise the exposure to halogenated compounds, known for their toxic effects as most of POPs, during life stages of development. Therefore, human matrices testifying of exposure occurring during pregnancy (human milk, placenta, meconium) were studied. In order to extend to the rest of the population, adipose tissue was also analysed. As previously detailed, latter generation of mass spectrometer are needed to reach sufficient selectivity. LC-HRMS was mainly used with the complementary GC- HRMS. Strategies were elaborated to process data leading to promising proof-of-concept and the identification of emerging biomarkers of exposure without a priori knowledge.

- 71/296 -

2. CHAPTER 2

INSTRUMENTAL METHOD DEVELOPMENT AND DATA PROCESSING TOOLS

- 73/296 -

Chapter 2. Instrumental method development and data processing tools

2.1. Introduction

The exhaustive characterisation of the chemical exposome is complex because of the diversity of existing compounds and the wide extent of compounds not yet discovered. The development of non-targeted methodological approaches is facing a number of conceptual and experimental challenges to characterise this exposome. It requires considering new analytical paradigms and workflows compared to conventional targeted methods. As previously detailed (Chapter 1), this paradigm shift takes place in the three main steps of the workflow, namely sample preparation, analysis and data processing. The development of an appropriate sample preparation procedure appears as the first step of such analytical pipeline, to optimise the compatibility of the resulting sample extract with the measurement system to be used. This sample preparation also depends on the complexity of the matrix (further detailed in Chapter 3). However, in practice, the testing and evaluation of various technical options first required a suitable analytical system and data processing capacity in order to generate the necessary signal to be interpreted. In the present chapter, this aspect led to primarily consider the instrumental methodological development and associated data processing. The generated raw analytical data (in terms of file format and size, amount and type of recorded information) depends on the implemented instrumental method. In the present case of NTS method development to characterise the human chemical exposure, liquid or gas phase chromatography (LC or GC) coupled to high resolution mass spectrometry (HRMS) in full-scan mode was used to cover a large range of molecules. This accurate and powerful technology leads to large volume of generated data (thousands of peaks) with many unidentified compounds. Then, dedicated bioinformatics tools were required to process those raw data, allowing for the further identification of the underlying markers of exposure.

The present chapter deals with our methodological optimisation conducted both in LC- and GC-HRMS. Both instrumentations were used in order to seek molecules on a large range of polarity (polar/semi-polar to nonpolar) thanks to the complementarity of those technologies. In addition, the nature of the data generated is specific to each of these two instruments and the bioinformatics tools to be used are also unique. Therefore, the data processing step cannot be discussed separately from the analytical method. Consequently, this chapter also presents strategies developed during this Ph.D. project to interpret the data, as well as the underlying bioinformatics tools developed in collaboration with and by a team of computer scientists and some partners of the HBM4EU project.

- 75/296 - Chapter 2. Instrumental method development and data processing tools

Sought unknown compounds can be quite different from each other in terms of physicochemical properties, structure and chemical composition. Because of this diversity, data processing, aiming to extract the most useful information from the global generated data set, appears as a long and laborious task requiring strategical prioritisations. A special focus on halogenated compounds was defined for the present Ph.D. work. Indeed, it is already known that many hazardous organic contaminants are halogenated, such as persistent organic pollutants. In addition, chlorinated and brominated molecules have a particular fingerprint in high-resolution mass spectrometry and can be extracted from the data set with appropriate and already described (HaloSeeker) bioinformatics tools.

2.2. Instrumental method optimisation

2.2.1. QA/QC mix of standards

Non-targeted analytical methods aim to cover a large range of compounds with broad physicochemical properties in terms of detectability, at the expense of sensitivity and quantification performance. Consequently, generic instrumental settings are required to recover as many compounds as possible. To assess these analytical conditions, two mixes of analytical standards was elaborated to guide our process. The first one was constituted based on a previous study (Léon et al., 2019) where Cl/Br-phenolic compounds were detected in marine animals. This mix prepared at a concentration of 0.1 ng µL-1 is summarised in Table 2.1 and further named “standard mix 1” with “Cl/Br-phenolic compounds”

Table 2.1: List of reference standards used for method development and performance (further named “Cl/Br-phenolic compounds” constituting the “standard mix 1”).

Name Acronym Formula MI mass (Da) 2-Chlorophenol

3-Chlorophenol Cl-phenol C6H5ClO 126.99561 4-Chlorophenol 2.3-Dichlorophenol 2.4-Dichlorophenol 2.5-Dichlorophenol DiCl-phenol C H Cl O 160.95664 2.6-Dichlorophenol 6 4 2 3.4-Dichlorophenol 3.5-Dichlorophenol 2.3.4-Trichlorophenol TriCl-phenol C H Cl O 194.91767 2.3.5-Trichlorophenol 6 3 3

- 76/296 - Chapter 2. Instrumental method development and data processing tools

2.3.6-Trichlorophenol 2.4.5-Trichlorophenol 2.4.6-Trichlorophenol 3.4.5-Trichlorophenol 2.3.4.5-Tetrachlorophenol

2.3.4.6-Tetrachlorophenol TetraCl-phenol C6H2Cl4O 228.87870 2.3.5.6-Tetrachlorophenol

Pentachlorophenol PentaCl-phenol C6HCl5O 262.83972 2-Bromophenol

3-Bromophenol Br-phenol C6H5BrO 170.94510 4-Bromophenol 2.3-Dibromophenol 2.4-Dibromophenol 2.5-Dibromophenol DiBr-phenol C H Br O 248.85561 2.6-Dibromophenol 6 4 2 3.4-Dibromophenol 3.5-Dibromophenol 2.3.4-Tribromophenol 2.3.5-Tribromophenol 2.3.6-Tribromophenol TriBr-phenol C6H3Br3O 326.76612 2.4.5-Tribromophenol 2.4.6-Tribromophenol 3.4.5-Tribromophenol 2.3.4.5-Tetrabromophenol

2.3.4.6-Tetrabromophenol TetraBr-phenol C6H2Br4O 404.67664 2.3.5.6-Tetrabromophenol

Pentabromophenol PentaBr-phenol C6HBr5O 482.58715

2’-hydroxy-2.4.4’-trichlorodiphenyl ether HtCDE (triclosan) C12H7Cl3O2 286.94388

2’-Hydroxy-2.4.4’-tribromodiphenyl ether HtBDE C12H7Br3O2 418.79234 6-Hydroxy-3-Chloro-2.2’.4.4’- HCteBDE C H Br ClO 530.66388 tetrabromodiphenyl ether 12 5 4 2 6-Hydroxy-2.2’.3.4.4’- HpBDE C H Br O 574.61336 pentabromodiphenyl ether 12 5 5 2 6-hydroxy-3.5-Dichloro-2.2’.4.4’- HdCteBDE C H Br Cl O 564.62491 tetrabromodiphenyl ether 12 4 4 2 2 6-Hydroxy-2.2’.3.4.4’.5- HhBDE C H Br O 652.52388 hexabromodiphenyl ether 12 4 6 2

We designed a second mix in order to encompass a larger range of known halogenated markers of exposure, thereby replacing the first mix for the subsequent experiments, since it was more suited to the objectives of this Ph.D. project. It is composed of halogenated contaminants of various molecular mass and polarity, with compounds used for different applications (pesticides, flame retardants) or with different health effects (endocrine disruptors) in their

- 77/296 - Chapter 2. Instrumental method development and data processing tools native or metabolised forms. A mix of 30 chlorinated and brominated compounds was constituted for this method development and performance assessment (Table 2.2 and Appendix 2) and was named “QA/QC mix 2”. Compounds were selected based on literature (López-Blanco et al. 2016), exchanges between HBM4EU WP16 partners, and our own individual expert view including pre-existing knowledge with regard to human chemical exposure gained by LABERCA and also necessary practical considerations (e.g. commercial availability). Moreover, these compounds are detectable in LC-ESI(+/-)-MS and/or in GC-EI-MS (Figure 2.1). They were diluted in two stock solutions (1 ng µL-1), in acetonitrile for standards supplied in methanol or acetonitrile, and in toluene for standards supplied in toluene, cyclohexane or isooctane. Then, both solutions were combined and diluted in acetonitrile (0.1 ng µL-1).

Table 2.2: List of the reference standards which substituted the standard mix 1 with Cl/Br-phenolic compounds and used for method development and performance assessment (further named “QA/QC mix 2”).

MI mass No. Name Acronym Formula log P (Da)

1 2.4-Dichlorophenol 2.4-DCP C6H4Cl2O 161.96392 3.1

2 3.5.6-Trichloro-2-pyridinol TCPy C5H2Cl3NO 196.92020 3.2

3 Simazine - C7H12ClN5 201.07812 2.2

4 Fenvalerate free acid - C11H13ClO2 212.06041 3.4

5 2.3.4.5-Tetrachlorophenol 2.3.4.5-tetraCP C6H2Cl4O 229.88598 4.5

6 2.4-Dibromophenol 2.4-DBP C6H4Br2O 249.86289 3.2

7 Acetochlor - C14H20ClNO2 269.11826 3.2

8 Hexaclorobenzene HCB C6Cl6 281.81312 5.7

9 Metolachlor - C15H22ClNO2 283.13391 3.1

10 beta-Hexachloroclyclohexane β-HCH C6H6Cl6 287.86007 3.8

11 Triclosan - C12H7Cl3O2 287.95116 5.0

12 Fenhexamid - C14H17Cl2NO2 301.06363 4.4

13 p.p'-dichlorodiphenyldichloroethylene p.p'-DDE C14H8Cl4 315.93801 7.0

14 Chlorpyrifos - C9H11Cl3NO3PS 348.92628 5.3

15 Chlorfenvinphos - C12H14Cl3O4P 357.96953 3.1

16 Tetraconazole - C13H11Cl2F4N3O 371.02153 4.4

17 Quizalofop-ethyl - C19H17ClN2O4 372.08768 4.3

18 Prochloraz - C15H16Cl3N3O2 375.03081 4.6

19 (Z)-Dimethomorph - C21H22ClNO4 387.12374 3.9

20 2.3.4.5-Tetrabromophenol 2.3.4.5-tetraBP C6H2Br4O 405.68392 5.4

21 2.3.5.6-tetrabromo-p-xylene p-TBX C8H6Br4 417.72030 5.4

22 Fipronil - C12H4Cl2F6N4OS 435.93871 4.5

- 78/296 - Chapter 2. Instrumental method development and data processing tools

23 Deltamethrin - C22H19Br2NO3 502.97317 6.2

24 Tetrabromobisphenol A TBBPA C15H12Br4O2 539.75708 6.8

25 Hexabromobenzene HBBz C6Br6 545.51003 6.1

26 alpha-hexabromocyclododecane α-HBCDD C12H18Br6 635.65088 7.1

27 Pentabromodiphenyl ether 153 PBDE 153 C12H4Br6O 637.53624 7.6

28 anti-Dechlorane plus a-DP C18H12Cl12 647.72013 8.0 6-Hydroxy-2.2.3.4.4.5- 29 OH-BDE 137 C H Br O 653.53116 7.2 hexabromodiphenyl ether 12 4 6 2 1.2-Bis(2.4.6- 30 BTBPE C H Br O 681.56246 7.7 tribromophenoxy)ethane 14 8 6 2 MI mass: monoisotopic mass

Figure 2.1: Venn diagram illustrating the LC-ESI(+/-)-HRMS and GC-EI-HRMS complementarity to detect the selected range of standards of the QA/QC mix 2 used to guide the present methodological development (30 test reference compounds listed in Table 2.2 and Appendix 2) (A). Illustration of their different physicochemical properties: monoisotopic mass and log P (B). Blue, red and black triangles are compounds detected in LC-ESI(-)-HRMS, LC-ESI(+)-HRMS and GC-EI-HRMS, respectively. A combination of colour indicates dual mode of detection. Yellow (no. 16) dot is the compound detected by three modes.

- 79/296 - Chapter 2. Instrumental method development and data processing tools

2.2.2. LC-HRMS method development

A LC-HRMS method previously developed for non-targeted screenings of chemical contaminants in sentinel environmental compartments and marine animals (Léon et al., 2019) was used as a basis for the present methodological work related to human matrices. This LC-HRMS method was optimised on an UltiMate 3000 UHPLC system coupled to a high resolution mass spectrometer (Orbitrap) Q-Exactive™ equipped with heated electrospray ion source (HESI-II, Thermo Fisher Scientific, San José, CA, USA), a hybrid quadrupole Orbitrap mass spectrometer with an HCD cell. External calibration was performed by infusing calibration mixture for negative and positive ionisation mode (MSCAL6 and MSCAL5 ProteoMass LTQ/FT-Hybrid, Supelco, Bellefonte, PA, USA). The instrument was controlled by Xcalibur (ThermoFisher Scientific) software version 3.0.

A reverse phase column (Hypersil Gold, Thermo Fisher Scientific) with a generic elution gradient of water and acetonitrile both supplemented with 10 mM of ammonium acetate was compared with three other stationary phases and two other solvent modifiers. The elution gradient duration and composition was optimised to ensure the reliability of the system during long batches. For all experiments, flow rate was maintained at 0.4 mL min-1 and oven temperature at 45 °C. In parallel, ion source temperature and S-lens frequency were compared to other values and finally kept at original MS settings. The Orbitrap device was used in full-scan acquisition mode in order to cover a large range of molecules at the maximum mass resolution (140 000 at 200 m/z).

LC method development

 Stationary phase optimisation

The initial stationary phase was a Hypersil Gold column (100 mm × 2.1 mm, 1.9 μm, Thermo Fisher Scientific). Its performance (mainly resolution of peak and chromatographic separation) was compared with three other reverse phase columns, Acquity BEH C18 (100 mm × 2.1 mm,

1.7 μm, Waters), Acquity CSH C18 (150 mm × 1.0 mm, 1.7 μm, Waters), Hypersil gold PFP (100 mm × 2.1 mm, 1.9 μm, Thermo Fisher Scientific). As shown in Figure 2.2, with Hypersil

Gold and Acquity BEH C18 columns compounds were well separated with symmetric peak shape and sufficient resolution. The two others stationary phases tested (Acquity CSH C18, Hypersil gold PFP) were found less satisfactory in terms of performance, hence they were not

- 80/296 - Chapter 2. Instrumental method development and data processing tools further investigated. In addition, as Hypersil Gold column was already largely used in the laboratory for different applications with a well-established robustness, it was eventually selected for the present work.

Figure 2.2: Extracted ion chromatograms by LC-ESI(-)-HRMS for 6 tribromophenol isomers (left) and α- and γ-HBCD (right), obtained after chromatographic separation on four

stationary phases namely Hypersil Gold, Acquity BEH C18, Hypersil Gold PFP and Acquity

CSH C18 (top to bottom).

 Mobile phase optimisation

Solvent modifier added in the mobile phase can improve peak shape, retention and ionisation ratio. For the present study, water and acetonitrile were both supplemented with ammonium acetate (10 mM), ammonium fluoride (0.2 and 1 mM) or acetic acid (0.1%). Ammonium fluoride is less common and should be used with caution to avoid fluoric acid formation. However, according to recent studies (Narduzzi et al., 2019), it can improve the ionisation in ESI of a wide range of compounds in comparison to ammonium acetate. The standard mix 1 of Cl/Br-phenolic compounds and the same extract of placenta spiked with those pure analytical standards were injected with each chromatographic condition after a sufficient time to equilibrate the column.

It resulted in significantly different total ion currents (TIC) for placenta sample extract with the various solvent modifiers tested (Figure 2.3). In particular, as peaks were insufficiently resolved with ammonium fluoride, the latter does not seem to be adapted for the present application.

- 81/296 - Chapter 2. Instrumental method development and data processing tools

Figure 2.3: Total ion currents obtained by LC-ESI(-)-HRMS for placenta sample extract eluted on Hypersil Gold column with water and acetonitrile supplemented with ammonium acetate 10 mM, ammonium fluoride 0.2 and 1 mM and acetic acid 0.1 % (top to bottom).

Similar TIC profiles were obtained between ammonium acetate and acetic acid. However, the retention of hydroxylated compounds is influenced by pH because of the OH ↔ O- form equilibrium of the alcoholic group. Ionic species were not retained on the stationary phase, therefore isomers separation was not efficient with the acetic acid modifier. This phenomenon was observed for tri- and tetrabromophenols and tri- and tetrachlorophenols (Figure 2.4). In the context of the present study where semi-polar to polar compounds are sought, a neutral pH with 10 mM of ammonium acetate was finally selected in order to retain as much compounds as possible.

- 82/296 - Chapter 2. Instrumental method development and data processing tools

Figure 2.4: Extraction ion chromatograms obtained for 3 tetrachlorophenol isomers after separation on Hypersil Gold column with water and acetonitrile supplemented with ammonium acetate 10 mM (top) and acetic acid 0.1 % (bottom)

 Elution gradient optimisation

The initial elution gradient (G1) was a mobile phase consisting of water (A) and acetonitrile (B), both supplemented with 10 mM ammonium acetate. The gradient started with A/B 80:20 (v/v) for 2 min, then linearly ramped to 20:80 within 5.5 min, and to 100% B over 6.5 min to be further maintained for 6 min. It was then returned to the initial conditions within 2 min and held for 4 min for a total run time of 26 min.

Alternative longer (G2) and more linear (G3) gradients were also tested to improve the chromatographic resolution. The QA/QC mix 2 was injected once with each gradient after a sufficient time of column conditioning. As illustrated in the Figure 2.5, compounds detected in ESI(-) appeared better separated with gradient G2 and G3 than G1. However, gradient G2 was twice longer than G3 for similar chromatographic resolution, so G3 was finally deemed as a good compromise to perform sufficient compound separation in a reasonable time.

- 83/296 - Chapter 2. Instrumental method development and data processing tools

Figure 2.5: Extracted ion chromatograms obtained for standards of the QA/QC mix 2 after separation on Hypersil Gold column with the three tested elutions gradients G1, G2 and G3.

As introduced in Chapter 1, sample preparation typically applied for non-targeted analyses is as minimal as possible in order to preserve all exogenous compounds of interest. However, matrix interferences (including lipids and proteins) remain in the final extract because of the limited purification. Consequently, the extract is more complex than a sample treated with a targeted sample preparation and advanced precautions should be taken. In order to ensure a good repeatability in retention time, which is an important factor for features and directly influences the data processing, a good reliability of chromatographic peak shape and column pressure over the batch are required. Thus, we implemented a column cleaning step after each injection. In practice, following the solvent gradient to elute compounds, a binary solvent acetone/isopropanol 1:1 (v/v) flushed the column and went to the waste, before a return to initial gradient conditions. This should be carefully configured (passing from a solvent to another one) to avoid irreversible damage for the column (Figure 2.6). In addition to this flush step and to prevent cross-contamination, the injection needle is rinsed before and after each injection (200

- 84/296 - Chapter 2. Instrumental method development and data processing tools

µL at 8 µL s-1) with a quaternary wash solvent water/methanol/acetonitrile/isopropanol in equal proportion.

Figure 2.6: LC-ESI(-)-HRMS total ion currents (left) obtained for a placenta sample extract and extracted ion chromatograms in the same sample of TBBPA and α-HBCD (right) at the beginning and the end of the batch with insufficiently optimised column flush.

 Conclusion

Chromatographic separation is a crucial step in non-targeted LC-MS analyses that directly impacts the further data processing step and compound identification through the helpful information related to the retention time. It is established that a non-targeted screening LC method is not as optimised as a targeted method because generic conditions suitable for many different compounds are sought. Several methods with similar performance could be developed, opening the window of possibilities in terms of optimisation. For the present study, parameters related to stationary and mobile phases, elution gradient and maintenance step have been optimised and selected settings are summarised in Table 2.3. The optimisation was run with the QA/QC mix 2 in LC-ESI(-)-HRMS and the same chromatographic conditions were selected for LC-ESI(+)-HRMS analysis. The compounds could be eluted in 15 min with symmetric peak shape.

The purpose of the QA/QC mix 2 is not only to assess method performance but also its limitations. Peak of polar molecules with an alcoholic group in the structure, eluted at the beginning of the gradient (before 8 min) were found less symmetric than others. HILIC could offer good complementarity to reverse phase to screen a larger range of polar to very polar compounds, with a mobile phase more compatible with MS detection than very aqueous mobile phase used with normal phase column. It is an interesting perspective of development specially

- 85/296 - Chapter 2. Instrumental method development and data processing tools to screen polar contaminant such as metabolites. Finally, non-targeted sample preparation leads to limited purified extract and supplementary preventive maintenance actions are required to ensure sufficient reliability over the batch of injections. In particular, the addition of a third solvent to flush the column after each injection appears necessary.

Table 2.3: LC settings selected after method optimisation.

Stationary phase Hypersil Gold (100 mm × 2.1 mm, 1.9 μm)

(A) Water + 10 mM ammonium acetate Mobile phase (B) Acetonitrile + 10 mM ammonium acetate (C) Acetonitrile/isopropanol 1/1 (v/v)

Gradient

Volume of injection 5 µL

Water/acetonitrile/methanol/isopropanol 1/1/1/1 (v/v/v/v) Needle wash Before and after each injection, occurring during column equilibration

Oven temperature 45 °C

Solvent flow 0.4 mL min-1

MS method development

 ESI source optimisation

Ionisation settings should be optimised with regards to the solvent flow rate, itself depending on the column used. The source temperature should allow sufficient desolvation for ionisation while preserving compounds from thermic degradation or in-source fragmentation. Once

- 86/296 - Chapter 2. Instrumental method development and data processing tools ionised, ions are transferred into the mass spectrometer through lenses and applied voltages should be suited to guide different ions. Indeed, excessively low or high voltages could reduce the focalisation and transmission of the ion beam and diminish the detected intensity.

Previously optimised chromatographic parameters (see Table 2.3) were first combined with ion source parameters established in a previous study from our research group (Omer et al., 2018). Then, we more specifically optimised the ion source temperature and the S-lens voltage using the QA/QC mix 2 (0.5 ng loaded on the column). First, temperatures of 150, 250 and 350 °C were tested with S-lens voltage at 50 arbitrary unit (AU). Then S-lens voltages at 30, 50, 70 and 100 AU were assessed using 350°C as the heater temperature, in positive and negative mode with a spray voltage of 3.5 and -2.5 kV, respectively. For both ionisation modes higher is the ion source temperature, higher is the signal intensity, except for acetochlor and chlorfenvinphos (Figure 2.7). The temperature of 350 °C was finally selected for our application. regarding S- lens voltage, 30 AU led to the lowest detected intensity for almost all compounds. The difference between 50, 70 and 100 AU is lower than a two-fold in detected intensity (Figure 2.8). As a compromise for most of compounds, S-lens voltage was set at 70 AU.

Figure 2.7: Optimisation of the ion source temperature. Detected intensity for compounds of QA/QC mix 2 in LC-ESI(-) (left) and LC-ESI(+) (right) for three temperature of 150, 250 and 350 °C with S-lens of 50 AU.

- 87/296 - Chapter 2. Instrumental method development and data processing tools

Figure 2.8: Optimisation of ion source S-lens value. Detected intensity for compounds of QA/QC mix 2 in LC-ESI(-) (left) and LC-ESI(+) (right) for a source temperature of 350 °C and four S-lens values of 30, 50, 70 and 100 AU.

 Orbitrap settings

The Orbitrap analyser offers high performance in terms of mass accuracy and resolution. The highest resolution reachable with the used instrument is 140 000 at 200 m/z (mass-to-charge ratio) and was set as such for all experiments. At this value, compounds at concentration 0.1 ng/µL (0.5 ng loaded on the column) were detected with more than 10 scan per peak and a mass accuracy lower than 1 ppm.

In order to cover a large range of molecules, acquisition mode in full-scan was used with two mass ranges, 100-1 000 and 1 000-2 000 Da, acquired in two separated injections. Most of our compounds of interest were expected in the first range, the second one was applied to confirm the compound identity by adduct detection as dimers (e.g. [2M+Cl]- adduct of HBCDD at 1306.27116 m/z).

- 88/296 - Chapter 2. Instrumental method development and data processing tools

 Conclusion

MS settings for non-targeted screenings required an HRMS instrument as the Orbitrap analyser. Just like for LC parameters, generic settings are required to cover a large range of molecules. The MS method used to analyse samples is detailed in Table 2.4 and Table 2.5. Both ionisation modes, positive and negative, were used to cover a broad range of molecules with the negative mode as first intention to mostly detect chlorinated and brominated compounds. Samples were analysed with both ionisation modes in two separated injections to ensure a sufficient amount of scan per peak.

Table 2.4: ESI source settings selected after method optimisation

Heater temperature 350 °C Capillary temperature 350 °C Sheath gas flow 50 AU Auxiliary gas flow 10 AU Sweep gas flow 0 AU S-lens radio frequency 70 AU 3.5 kV positive mode Spray voltages -2.5 kV negative mode

Table 2.5: HRMS settings selected after method optimisation.

Resolution 140,000 FWHM* at m/z 200 Mass-to-charge (m/z) range 100-1 000 or 1 000-2 000 AGC** target 5×105 Maximum injection time 500 ms Ion mode Positive or negative Acquisition mode Full-scan *: full width half maximum

**: Automatic gain control

- 89/296 - Chapter 2. Instrumental method development and data processing tools

2.2.3. GC-HRMS method development

A GC-HRMS method was developed in order to cover the complementary range of nonpolar molecules not detectable by LC-HRMS.The GC-Orbitrap coupling is marketed since 2015 (Peterson et al., 2014-Part I; Peterson et al., 2014-Part II) and was installed at LABERCA in July 2019. Due to the recent implementation in the laboratory, less work has been achieved on this instrument compared to the previously described LC-HRMS work.

GC-HRMS method was optimised on GC-Q-Orbitrap system (Q Exactive GC, Thermo Scientific, Bremen, Germany) consisting of a TriPlus RSH autosampler, a TRACE 1310 GC with a hot split/splitless injector, an electron ionisation (EI) source and a hybrid quadrupole Orbitrap mass spectrometer with an HCD cell. The spectrometer was tuned and calibrated externally. Instrument was controlled by Xcalibur software version 4.1.

As for the developed LC-HRMS method, a large range of molecules was sought and generic conditions were required. A nonpolar stationary phase (DB5-MS column, 30 m × 0.25 mm, 0.25 µm) was used to screen nonpolar compounds as a complementarity to those typically detected through LC-HRMS. Various GC parameters were optimised, namely solvent of injection and injector and oven temperatures. HRMS settings were comparable to those applied for LC-HRMS detection.

GC method development

 Injector and injection solvent optimisation

First of all, two injection solvents, namely hexane and toluene, were compared to analyse standards of the QA/QC mix 2. These two solvents are commonly used in GC and the main difference is their boiling point, 68 and 110 °C, respectively. Hexane makes possible to start the gradient of elution at lower temperature in order to extend the range of detectable molecules. Toluene is less volatile than hexane and can preserve compounds from evaporation during the sample preparation until the injection. The QA/QC mix 2 initially diluted in acetonitrile was dried under nitrogen stream and reconstituted in pure hexane or toluene. Both extracts were analysed with the same chromatographic and mass spectrometric conditions, except the initial oven temperature was set at 50°C and 90°C for hexane and toluene respectively. As illustrated in Figure 2.9, peak fronting is observed for compounds injected in toluene whereas the peaks are symmetric with hexane. A particularity of the present instrument is the top of the injector

- 90/296 - Chapter 2. Instrumental method development and data processing tools set at room temperature to facilitate maintenance actions. Our hypothesis is that a gradient of temperature occurred in the liner between the top at room temperature and the inside at 300 °C. This could influence the vaporisation phenomenon and the focalisation of the extract on column head leading to peak fronting. Hexane was finally selected to reconstitute the extract before injection in order to cover a wider range of molecules than toluene and ensure good peak shape. In order to enhance the vaporisation of toluene, the temperature of the injector was increased from 300 to 320 °C. This temperature was also set for injection of hexane.

Figure 2.9: GC-EI-HRMS extracted ion chromatogram obtained for acetochlor, metolachlor and p-TBX injected in toluene (left) and hexane (right).

 External contamination assessment and management

The vial septum might release different substances further detected during sample analysis. In order to characterise external contamination coming from the septum and not from the solvent, empty vial closed by screw cap with septum made of silicone/PTFE (polytetrafluoroethylene) and of natural rubber were compared. Figure 2.10 illustrates this external contamination released by natural rubber septum, whereas silicon/PTFE septum was inert. Even if silicon/PTFE septa were selected for injections, interfering peaks were detected when the same vial of hexane was injected more than once. Mass spectra of those peaks were matched against NIST library and were identified as siloxane compounds released by septum. Hexane purity was also checked by analysing the solvent stored in a vial covered by aluminium foil and none of those peaks were detected. Therefore, this external contamination cannot be avoided unless vial caps are removed before injection, which is not feasible when long batches run overnight. Alternatively, a possible solution to this issue is for users to ensure sufficient repeatability of the contamination, so that it can be taken into account during the data processing. The phenomenon mainly occurred during the second injection of the vial, so at the beginning of

- 91/296 - Chapter 2. Instrumental method development and data processing tools each sequence, two successive injections of solvent are carried out in order to acquire the signal of septum released compounds. Once the external contamination is characterised, or least is repeatable, it can thus be managed during the data processing.

Figure 2.10: Total ion currents obtained for empty vial closed by silicon/PTFE (black) or natural rubber (red) septum or aluminium foil (blue) to compare external contamination released by septum (left). Total ion currents obtained for vial containing hexane closed by silicon/PTFE septum injected twice with comparison to vial containing hexane closed by aluminium foil (right). Analyses were performed in GC-EI-HRMS.

 Elution gradient optimisation

Just as for the previous optimisation of the LC gradient of elution, generic settings were sought to efficiently resolve compounds in GC in a reasonable time of run. The QA/QC mix 2 was first injected in hexane with a linear gradient of 10°C min-1 from 60 to 320 °C. Then, another gradient including a slope at 5°C min-1 from 130 to 250 °C was tested in order to perform a better chromatographic separation. The resulting chromatographic separations were found very similar. However, the second gradient provided better resolution for compounds which were almost co-eluted with the linear gradient. In order to ensure sufficient separation for a large range of molecules this second gradient was selected. Eventually, the initial temperature was set at 60 °C for 2 min and increased to 130 °C at 10 °C/min, then to 250 °C at 5 °C/min and to 320 °C at 10 °C/min and held at 320 °C for 10 min. The total run time was 50 min and an example of chromatogram can be found in Figure 2.11.

- 92/296 - Chapter 2. Instrumental method development and data processing tools

Figure 2.11: GC-EI-HRMS extracted ion chromatograms obtained for compounds of the QA/QC mix 2 eluted with the following gradient of temperature: 60 °C for 2 min, increased to 130 °C at 10 °C/min, then to 250 °C at 5 °C/min and to 320 °C at 10 °C/min and held at 320 °C for 10 min.

Conclusion on GC-MS method development

Our GC-HRMS optimisation work was conducted within the same perspective as the previous LC-HRMS method, knowing that generic chromatographic and MS settings were sought in order to cover a wide range of various molecules. Finally selected GC and MS settings are summarised in Table 2.6 and Table 2.7, respectively.

As previously explained in Chapter 1, retention times in GC are comparable between different instruments and methods by using retention index (Ri). For the present study a saturated alkane mix C7-C40 was injected at the beginning of each batch to calculate Ri and facilitate the identification.

- 93/296 - Chapter 2. Instrumental method development and data processing tools

Table 2.6: GC settings selected after method optimisation.

Stationary phase DB5-MS column (30 m × 0.25 mm, 0.25 µm)

Carrier gas Helium

Gradient

Volume of injection 2 µL

Needle wash 1- Toluene; 2- hexane after each injection

Injector 320 °C Temperatures Transfer line 320 °C

Gas flow 1 mL/min, constant flow

Table 2.7: HRMS settings selected after method optimisation.

Heater temperature 300 °C EI source voltage 70 eV Solvent delay 5 min Resolution 120,000 FWHM* at m/z 200 Mass-to-charge (m/z) range 50-750 AGC** target 5×105 Maximum injection time 500 ms

Ion mode Positive

Acquisition mode Full-scan

*: full width half maximum **: Automatic gain control

- 94/296 - Chapter 2. Instrumental method development and data processing tools

2.3. LC- and GC-HRMS method performance

2.3.1. Calibration curve

The instrumental linearity of the optimised LC- and GC-HRMS methods was assessed on a realistic concentration range to detect organic contaminant in human matrix. For LC, the QA/QC mix 2 at 0.1 ng µL-1 was diluted in acetonitrile at concentrations 0.01 and 0.001 ng µL- 1. 1 and 5 µL of each were injected to extend the range of compound amount loaded on the column in ascending order of concentration. For GC, a solution at 0.5 ng µL-1 was prepared from the two stock solutions at 1 ng µL-1 of QA/QC mix 2. Solvent was evaporated and compounds were reconstituted in hexane before being diluted in hexane at concentrations 0.1, 0.05, 0.01, 0.005 and 0.001 ng µL-1. 1 µL of each was injected in ascending order of concentration. (Table 2.8). The instrument linearity for compounds of the QA/QC mix 2 was considered acceptable as the coefficient of determination was higher than 0.98 for an extended range of concentration and compounds.

Table 2.8: Instrumental linearity (coefficient of determination R2) on a calibration curve from 0.001 to 0.5 ng µL-1 for the QA/QC mix 2 analysed both in LC-ESI(+/-) and GC-EI.

Linearity (R2) No. Compound name GC-EI LC-ESI(-) LC-ESI(+) 1 2.4-DCP 0.992* 0.981 - 2 TCPy - 0.986 - 3 Simazine 0.993* - 0.996 4 Fenvalerate free acid - 0.981 - 5 2.3.4.5-tetra-CP 0.982* 0.985 - 6 2.4-DBP 0.990* 0.987 - 7 Acetochlor 0.990 - 0.995 8 HCB 0.989 - - 9 Metolachlor 0.989 - 0.996 10 β-HCH 0.994 0.992* - 11 Triclosan - 0.989 - 12 Fenhexamid - 0.990 - 13 p,p'-DDE 0.986 - - 14 Chlorpyrifos 0.983 - - 15 Chlorfenvinphos 0.984 - 0.992 16 Tetraconazole 0.847 0.987 0.996 17 Quizalofop-p-ethyl NS - - 18 Prochloraz NS* - 0.996 19 (Z)-Dimetomorph NS* - 0.995 20 2.3.4.5-tetra-BP NS* 0.983 -

- 95/296 - Chapter 2. Instrumental method development and data processing tools

21 p-TBX 0.985 - - 22 Fipronil 0.747* 0.988 - 23 Deltametrin 0.982 24 TBBPA - 0.987 - 25 HBBz 0.984 - - 26 α-HBCDD - 0.988 27 PBDE 153 0.982 - - 28 a-DP 0.983 - - 29 OH-BDE 137 - 0.987 - 30 BTBPE 0.978 - - * compound usually detected with the other ionisation mode NS: Not Significant. Compound detected but only two points were considered for the calibration curve.

2.3.2. Instrumental sensitivity

The instrumental limits of detection (LOD) obtained with the optimised LC- and GC-HRMS methods were assessed using the same calibration curve as previously described to verify the linearity of the instrument. It was considered as the lowest concentration detecting a chromatographic peak defined as higher than the baseline for at least 5 consecutive scans at expected retention time. If a compound was detected according to these criteria at the lowest concentration, then the LOD was considered lower than 0.001 ng µL-1 (Table 2.9). The achieved instrumental limits of detection were considered acceptable to detected organic contaminants in human matrices for an extended range of concentration and compounds. It represents the best case without interference occurring during the ionisation. It was assumed that the sensitivity can be different in presence of matrix interference leading to signal enhancement or suppression. This matrix effect has to be assessed for each sample preparation and matrix.

Table 2.9: Instrumental limits of detection obtained in LC-ESI(+/-) and GC-EI for compounds of the QA/QC mix 2.

LOD (ng/µL) No Compound name GC-EI LC-ESI(-) LC-ESI(+) 1 2.4-DCP 0.1* 0.01 - 2 TCPy - 0.01 - 3 Simazine 0.01* - <0.001 4 Fenvalerate free acid - 0.01 - 5 2.3.4.5-tetra-CP 0.05* <0.001 - 6 2.4-DBP 0.5* 0.005 - 7 Acetochlor <0.001 - 0.01 8 HCB <0.001 - -

- 96/296 - Chapter 2. Instrumental method development and data processing tools

9 Metolachlor <0.001 - <0.001 10 β-HCH <0.001 0.05* - 11 Triclosan - <0.001 - 12 Fenhexamid - <0.001 - 13 p,p'-DDE <0.001 - - 14 Chlorpyrifos <0.001 - - 15 Chlorfenvinphos <0.001 - 0.005 16 Tetraconazole <0.001 <0.001 0.005 17 Quizalofop-p-ethyl 0.5 - - 18 Prochloraz 0.1* - <0.001 19 (Z)-Dimetomorph 0.1* - <0.001 20 2.3.4.5-tetra-BP 0.5* <0.001 - 21 p-TBX <0.001 - - 22 Fipronil 0.05* <0.001 - 23 Deltametrin 0.05 24 TBBPA - 0.005 - 25 HBBz <0.001 - - 26 α-HBCDD - 0.01 27 PBDE 153 0.005 - - 28 a-DP 0.005 - - 29 OH-BDE 137 - <0.001 - 30 BTBPE 0.01 - - *: compound usually detected with the other ionisation mode.

The complementarity of the two techniques allows to detect the thirty compounds of the QA/QC mix 2 within three ionisation modes. Many compounds were detected by GC-EI-HRMS while they are better detected by LC-ESI(+/-)-HRMS. For instance, β-HCH was detected in LC-ESI(-)-HRMS with acetate adduct while it is better ionised by GC-EI-HRMS. Based on the establishment of the instrumental LOD, compounds with values marked with asterisk in Table 2.9 were not considered with this ionisation mode for other investigations.

As expected, we observed that compounds with log P < 4 present higher intensities in LC-ESI(+/-)-HRMS whereas compounds with log P > 4 are better detected in GC-EI-HMRS. However, this observation is not a stringent rule as some compounds as acetochlor and metolachlor (log P 3.2 and 3.1, respectively) present the same and low LOD in both LC-ESI(+)-HRMS and GC-EI-HRMS. In the case of TBBPA and OH-BDE 137, those two compounds present log P > 6 and are not detected in GC. As their chemical structure is made of hydroxy group, it increases their ionisation in LC-ESI(-)-HRMS with hydrogen donor facilitating the formation of [M-H]- ion. Those hydroxylated compounds could be detected with derivative agent to hide this hydroxy function and to increase the volatility of the compounds. If this method is already well established for targeted analyses, it could make the data

- 97/296 - Chapter 2. Instrumental method development and data processing tools processing more complex of non-targeted screening analyses. Indeed, the fragmentation occurring during the ionisation provides fragments from the derivative agent, in addition to the characteristic fragments of the molecule of interest. As the identification is recognised as the bottleneck of the non-targeted strategies and more development are required to process GC- HRMS data, the derivatisation option was not retained for the present study. Moreover, it was assumed that the complementarity of LC- and GC-HRMS detection would be more efficient if the sample preparation was also developed in that sense by for instance partitioning more polar compounds from less ones in two fractions, and then analysed by LC- and GC-HRMS respectively. Thus, hydroxylated compounds would be preferentially expected in the polar fraction.

2.3.3. Instrument repeatability

Instrumental repeatability of the optimised LC- and GC-HRMS methods was assessed for compounds in the QA/QC mix 2 at the concentration 0.1 ng µL-1 in pure solvent. The same vial was injected four times during a batch of 40 injections of real human samples, at the beginning, twice in the middle and at the end. Relative standard deviation (RSD) of the measured signal intensities are reported in Table 2.10 for compounds previously prioritised based on the instrumental LOD. The repeatability was considered acceptable when RSD were lower than 30%. In the particular case where the RSD exceeds 30%, for instance quizalofop-p-ethyl (62%), the compound was considered as a limit of the method.

Table 2.10: Repeatability (relative standard deviation) observed in LC-ESI(+/-) and GC-EI for compounds of the QA/QC mix 2 at the concentration 0.1 ng µL-1.

RSD No Compound name GC-EI LC-ESI(-) LC-ESI(+) 1 2.4-DCP - 8% - 2 TCPy - 18% - 3 Simazine - - 2% 4 Fenvalerate free acid - 37% - 5 2.3.4.5-tetra-CP - 17% - 6 2.4-DBP - 9% - 7 Acetochlor 18% - 6% 8 HCB 21% - - 9 Metolachlor 18% - 8% 10 β-HCH 29% - - 11 Triclosan - 14% - 12 Fenhexamid - 9% -

- 98/296 - Chapter 2. Instrumental method development and data processing tools

13 p,p'-DDE 17% - - 14 Chlorpyrifos 14% - - 15 Chlorfenvinphos 33% - 3% 16 Tetraconazole 23% 17% 11% 17 Quizalofop-p-ethyl 62% - - 18 Prochloraz - - 5% 19 (Z)-Dimetomorph - - 5% 20 2.3.4.5-tetra-BP - 10% - 21 p-TBX 24% - - 22 Fipronil - 14% - 23 Deltametrin 27% 24 TBBPA - 16% - 25 HBBz 43% - - 26 α-HBCDD - 9% 27 PBDE 153 22% - - 28 a-DP 22% - - 29 OH-BDE 137 - 10% - 30 BTBPE 29% - -

2.3.4. Discussion

Performance of the developed LC- and GC-HRMS methods were assessed based on the linearity, LOD and repeatability of compounds from the QA/QC mix 2. As opposed to the targeted methods, it is not actually possible to validate the method as it will be used to seek other compounds not included in the QA/QC mix 2. However, according to the diversity of compounds in this mix, it is possible to draw some method limitations. For the present non-targeted analytical approach, the LC/GC complementarity enabled to detect all compounds from the QA/QC mix 2. However, two molecules, fenvalerate free acid (no. 4) detected by LC-ESI(-)-HRMS and quizalofop-p-ethyl (no. 17) detected by GC-EI-HRMS, were highlighted as method limitation because of high LOD (≥ 0.01 ng µL-1) and poor repeatability (RSD >30%). The next step is not to go deeper in the method optimisation for these two compounds, as it would probably cause a non-satisfying detection of some others molecules. However, it is more interesting to develop another non-targeted method able to detect these compounds at lower concentration. Moreover, those method limitations could be extended to other similar molecules with statistical approach. A concrete example of method assessment is detailed in Chapter 3 (section 3.4.3.5 Predicting the recovery through the physicochemical properties of the considered markers), based on compounds physicochemical properties and supported by a modelling approach.

- 99/296 - Chapter 2. Instrumental method development and data processing tools

Beyond the assessment of the method, it is crucial to ensure reproducibility over batches independently of the nature of analysed samples. For each LC and GC sequence of acquisition, the instrument was externally calibrated with a commercial mix and the QA/QC mix 2 was injected at least twice to verify the instrument stability. Intensities were reported on a control chart to both assess the instrument sensitivity and follow its evolution over months. This QA/QC mix 2 was injected during the sequence in pure solvent or contained in QC sample, to control the mass accuracy. In addition, real samples were spiked with labelled external standards, from which, the signals were integrated and the retention time and peak shape were verified. If the external standard was not detected, the sample was re-injected. Once the sequence had been validated, the data processing can start with appropriate software.

2.4. Data processing and bioinformatics tools developed for non-targeted screening

2.4.1. Recently developed software to process non-targeted data generated by HRMS instruments

Data processing applied to targeted approaches is generally based on the detection and quantification of a priori known compounds with already recorded reference data related to their chemical structure and detection characteristics stored in library or home-made database. These data are commonly acquired in low-resolution mass spectrometry (LRMS) and many vendor or in-house software already exist to process it. Processing of non-targeted data acquired from full-scan detection mode is more challenging because unknown exogenous contaminants are sought in a forest of peaks of endogenous compounds. Moreover, the current lack of common database and reliable computational software able to deal with such big amount of complex data (i.e. thousands of peaks) impair the investigation. LC- and GC- HRMS analyses typically generate data-sets containing thousands of features characterised by a given m/z, retention time and signal intensity. However, LC-ESI(+/-)-HRMS and GC-EI-HRMS do not provide the same information, not only because different molecules are detected with both instruments, but mainly because the ionisation is different. The low energy ESI source enhances the detection of molecular ions [M+H]+ and [M-H]- and adducts, whereas the highly energetic EI source increases molecule fragmentation. From a comparative point of view, GC-EI spectra are more structurally informative due to the greater fragmentation and are also more reproducible. Since all data are acquired via a single-channel, EI spectra are analogous to

- 100/296 - Chapter 2. Instrumental method development and data processing tools

LC-MS1 spectra with very high in-source fragmentation, whereas LC-ESI MS/MS are collected on two distinct channels (MS1 and MS2). Consequently, both LC and GC data were separately investigated to screen halogenated compounds and the associated strategies are described below.

Methodology to screen for halogenated compounds by HRMS

As previously mentioned, a special focus on chlorinated and brominated compounds was made during this Ph.D. project. Such compounds exhibit a particular fingerprint in mass spectrometry thanks to their natural isotopic pattern. In the case of chlorinated and brominated atoms, there are indeed two stable isotopes with significant natural abundances of 0.7578/0.2422 for 35Cl/37Cl and 0.5069/0.4931 for 79Br/81Br. This intrinsic criterion distinguishes those atoms from others commonly find in organic molecules (C, H, O, N). It leads to characteristic isotopic patterns following laws of probability depending on the number of Cl or Br atom (n) constituting the ion and the natural isotopic abundance. Theoretically, there are n+1 peaks in the resulting isotopic pattern (Figure 2.12). This specificity known for halogenated ions allows for their straightforward recognition in a mass spectrum, either visually or thanks to bioinformatics tools.

Figure 2.12: Theoretical isotopic patterns resulting from 1 to 10 chlorinated (left) and brominated (right) atoms in mass spectrometry.

According to the international union of pure and applied chemistry (IUPAC) system, the 12C atom exact mass is set at 12.00000 u.m.a. and serves as reference for the other elements. In chemistry and by misuse of language, the mass defect is the difference with the nearest unit (value between -0.5 and +0.5). Moreover, Cl and Br atoms have similar mass defects which significantly differ from other atoms C, H, O, N. To plot the mass defect, it appears more

- 101/296 - Chapter 2. Instrumental method development and data processing tools convenient to adapt the IUPAC system in order to benefit from this distinction and better visualise isotopic patterns. It was first suggested by Kendrick in 1963 for petrochemical applications employing CH2 exact mass being 14.0000 u.m.a. as the reference (Kendrick, 1963). Later, in 2010, Taguchi et al. proposed, as another reference, the substitution of hydrogen by 35Cl, 34.00000 u.m.a. (Taguchi et al., 2010). This way, polychlorinated series exhibit the same mass defect and are more easily recognised, since they appear on a horizontal line on “mass defect-plots” (MD plot). Conveniently, polybrominated series are also aligned on a horizontal line.

Figure 2.13: MD plot obtained from HaloSeeker v.1.0 with clusters of mono- to penta- chlorophenols (in rectangles) and mono- to penta-bromophenols (in circles) horizontally aligned.

The conversion from the IUPAC to the H/Cl-scale is detailed in the following formula,

34,00000 푚 = 푚 ∗ 푢. 푚. 푎. 퐻/퐶푙 퐼푈푃퐴퐶 34.96885 − 1.00783 where the denominator represents the IUPAC masses of 35Cl and 1H.

This approach based on isotopic pattern and mass defect can be used to evidence chlorinated and brominated compounds. This complex data process, historically performed manually within our research group with VGA script, was a matter of semi-automation through the development of an advanced software named HaloSeeker specifically designed for this purpose.

HaloSeeker software

The application HaloSeeker-version 1 (Léon et al., 2019) was developed in 2019 by researchers and bioinformatics scientists at LABERCA to address the issue of non-targeted screening of halogenated compounds detected by HRMS. This user-friendly application helps investigating

- 102/296 - Chapter 2. Instrumental method development and data processing tools chlorine- and bromine-containing compounds based on H/Cl-scale mass defect. It is available on request at [email protected].

On the back end, HaloSeeker runs under the R environment (open access licence). The workflow includes three main steps (Figure 2.14). First, data files (for instance, .raw data generated by Thermo instrument) are converted into mzXML universal format with MSConvert script. Second, the peak-picking step uses centWave algorithm from xcms R package to detect symmetric chromatographic peaks. It highlights regions of interest with similar m/z in restricted rT ranges (i.e. consecutive MS scans). It results in a table of so-called features characterised by m/z, rT and intensity. Third, features are paired in clusters according to the isotopic mass differences of C, Cl and Br isotopes. More precisely, in a cluster, each feature is named as an isotopologue when two features defer from one to another solely by the presence of one or more isotopes (12/13C, 35/37Cl, 79/81Br). A Cl- or Br-containing cluster is composed of at least four isotopologues A; A+1; A+2; A+3 (Figure 2.14 and Figure 2.15).

Figure 2.14: Synoptic view of HaloSeeker data processing workflow in the back end.

On the front end, each cluster is plotted on a H/Cl-scale MD plot with the fractional part on the y-axis as a function of the precise mass on the x-axis. Exact mass differences between Cl or Br isotopes being slightly different (Table 2.11), it is possible to take advantage of the slope formed by alignment of feature dots from a cluster, to hypothesise the presence of Cl (negative slope) and/or Br (positive slope) (Figure 2.15).

- 103/296 - Chapter 2. Instrumental method development and data processing tools

Table 2.11: Exact masses of chlorinated and brominated isotopes.

35Cl 37Cl 79Br 81Br Exact IUPAC mass 34.96885 36.96590 78.91834 80.91629 Difference (IUPAC) 1.99705 1.99795 Difference (H/Cl scale) 1.99934 2.00024

Figure 2.15: Synthetic view of a H/Cl-scale MD plot (left) and details of two Cl-containing homologue clusters (right). Each dot represents a given features characterised by a given m/z, retention time and intensity values; group of paired features, or isotopologue, is defined as a cluster.

Several functions have been implemented in the application to annotate and prioritise signals of interest:

o Dereplication compares experimental data with a database loaded in HaloSeeker. The scoring is based on pattern similarities. o Formula decomposition is available to assign chemical formula with a confidence level according to available information. o Many display options are available for the MD plot: filters according to rT, m/z and intensity ranges (Table 2.12), filters on features paired after pairing (F1) or passing further ion ratio rules specific to halogenated clusters (F2) and polyhalogenated clusters (F2+). In order to perform signal prioritisation, HaloSeeker uses two main parameters namely the exact mass (and retention time) and the signal intensity. First, paired clusters are discriminated from non-paired clusters when an A+2 isotopologue matches with the exact mass (substitution of

- 104/296 - Chapter 2. Instrumental method development and data processing tools

35Cl by 37Cl, or 79Br by 81Br), thanks to the filter F1. Then, clusters are filtered according to ion ratio between isotopologues, depending on implemented rules detailed in Table 2.12 with filters F2 and F2+.

Table 2.12: HaloSeeker filter rules to prioritise signals of interest from peak-picked features to polyhalogenated chemical species.

F0 All peak-picked features F1 m/z-paired clusters (i.e. containing at least A+2 isotopologue) F1 rules AND A-2 ratio = 0 (absence) AND A+2 ratio ≥25% F2 OR A-2 ratio ≥60% AND A+2 ratio ≥20% OR A-2 ratio ≥27% AND A+2 ratio ≥36% F2 rules AND F2+ Monohalogenated ions are removed according to: A-2 and A+4 ratios = 0 (absence) AND; A+2 ratio ∈ [25−39] ∪ [77−117]%

Thanks to HaloSeeker, data are prioritised to highlight halogenated molecules and thus create clusters. Then, a second application named AlignDataShiny was developed and includes: o An alignment module, which adjusts cluster retention times and groups clusters detected in several samples over a batch. This step has already been implemented in different software, for instance Galaxy (Giardine et al., 2005) which is used for “omics” applications. In the present case focusing on halogenated compounds, the difficulty remains in the alignment of clusters (group of features) instead of single features because of the orbitrap error related to exact peak intensity (Appendix 3). o Procedural blank sample subtraction or suppression, as a way to prioritise signals of interest in order to only focus on clusters detected in samples. In parallel to the first version of HaloSeeker, AlignDataShiny was developed and used during the Ph.D. project pending the development of HaloSeeker-version 2, which includes these options since June 2020.

MS-DIAL software

MS-DIAL software was firstly used and investigated during a collaboration established with WP16 partners at the Research Centre for Toxic Compounds in the Environment (RECETOX,

- 105/296 - Chapter 2. Instrumental method development and data processing tools

Czech Republic) as part of the HBM4EU project. This collaboration was conducted through Ph.D. student secondment, including my own stay during nine weeks in winter 2020 at RECETOX under the supervision of Pf. Jana Klanova. Due to its very recent implementation in LABERCA, less data files were processed with this software compared to HaloSeeker. However, it appeared as a software of interest, in particular for processing GC-HRMS data.

MS-DIAL was initially developed to process LC-MS/MS-DIA data for the identification of small molecules by mass spectral deconvolution (Tsugawa et al., 2015). Spectral deconvolution is commonly applied for GC-EI data and MS-DIAL later incorporated functions for processing GC-MS datasets. In combination with MS-FINDER (Lai et al., 2018), another software developed by the same team, it is possible to query the NIST library (National Institute of Standards and Technology). However, GC-LRMS data are mainly recorded in NIST library and a new database for GC-HRMS is needed.

HBM4EU database

As previously detailed in the section 1.4.4.2. Compound annotation, a database containing all known and suspect CECs, including metabolites with harmonised information does not yet exist in the exposomics field. However, this database appears as a crucial tool to annotate non-targeted data and to highlight signals of interest for investigation. This issue has been an important task of the HBM4EU-WP16 project since its launch. First, several already existing databases from various platforms such as NORMAN network, Mass bank collection, US EPA dashboard, etc. were merged (article written by Meijer et al., is in progress). Then, the resulting list containing 145 284 entries was curated to reduce to 70 583 unique compounds in terms of both structural and stereochemistry. Finally, generation of experimental data are still required to consolidate and transform this vast list in a strong library. Also, in silico methods and other approach will be used to generate metabolite data, based on parent compounds included in the list.

This list includes all primary information such as name, chemical formula, InChIKey, SMILE, exact mass, etc. As a first intention and for all compounds, it is possible to calculate adduct + + - - mass detected by LC-ESI, for instance [M+H] , [M+Na] and [M-H] , [M+CH3COO] in positive and negative ion mode, respectively. For the present Ph.D. thesis work, this list named the “HBM4EU list” was used to putatively identify compounds detected in human samples by LC-ESI-HRMS. An advanced version of this list is needed to process GC-EI-HRMS data. In

- 106/296 - Chapter 2. Instrumental method development and data processing tools fact, adducts can be calculated because the mass addition/loss to the ion is known, but in GC- EI-HRMS, the fragmentation spectrum is almost unique for each compound. Therefore, a real library including experimental or in-silico spectral fragmentation is still needed. Thus, the HBM4EU list is not directly applicable to process GC-HRMS acquired during the present Ph.D. project.

2.4.2. Strategy developed to process data generated by non-targeted approaches

Non-targeted data processing can be conducted by different ways and software. According to the previous section describing tools used during the Ph.D. project, the present section details the strategy developed to implement those approaches in the processing of LC- and GC-HRMS data and perspectives based on the gained experience.

Investigation of LC-HRMS data

 Developed and adopted strategy

During the present Ph.D. work, the strategy to process LC-HRMS data was established step-by-step over 2018 and 2019 (Figure 2.17). When HaloSeeker was not required, for instance to assess method capabilities, Xcalibur was used to integrate chromatographic peaks, verify retention time and mass spectra standards of interest, with a mass tolerance of 5 ppm. For data investigation, compound annotation and even identification, the following strategy was used. First of all, raw data was converted within HaloSeeker to highlight chlorinated and brominated molecules. Default settings were tested on compounds of the QA/QC mix 2 and adopted, since they allowed to visualise all compounds and corresponding adducts (Figure 2.16). Peak-picking parameters were m/z tolerance, 5 ppm; peakwidth, 5-60 s; pre-filter step, 3; pre-filter level 10 000 and sntresh, 10. Halogen-pairing parameters were rT tolerance, 1 second and m/z tolerance, 0.5 mDa.

- 107/296 - Chapter 2. Instrumental method development and data processing tools

Figure 2.16: MD plot obtained by the HaloSeeker v1.0 software after the processing of typical LC-ESI(-)-HRMS data file acquired from a solvent sample containing standards of the QA/QC mix 2.

Then, all detected clusters were listed and saved in the database which was transferred into AlignDataShiny where they were aligned. Once matrix and blank samples were aligned, the result was exported in an excel file. Signals from procedural blank were manually discarded. As a first intention and in order to prioritise data, only signal detected in matrix samples and not in procedural blanks were investigated. Then, signals were manually classified based on their observed frequency of detection when it was applicable. Finally, the expertise of the user is also part of the process to estimate if the signal should be discarded (peak shape, intensity, halogenated pattern etc.). This prioritisation approach allowed to reduce by four the number of signals of interest (approximately 50 to 100 signals). For highlighted clusters, the monoisotopic mass was matched against the HBM4EU list to putatively annotate some signals. Since AlignDataShiny provides a list of base peak mass which is not systematically the monoisotopic peak, HaloSeeker was used in parallel to verify the cluster pattern. If the base peak was not the

- 108/296 - Chapter 2. Instrumental method development and data processing tools monoisotopic peak, the exact mass provided by HaloSeeker was used to query the HBM4EU list. If there was a hit with a compound and if the experimental mass spectrum and retention time were in accordance with this compound, then it was fragmented. MS/MS experiments were also conducted on pure analytical standards when available to validate the identification at level 1 according to the confidence scale proposed by Schymanski et al. (2014). The concept of annotation and identification is also a prominent part of NTS discussions. As detailed in the Chapter 1, different confidence level according to different scale exist and at the present time there is no consensus regarding this aspect.

Figure 2.17: Summary of the developed strategy to process non-targeted LC-HRMS data. *: step performed manually.

 Discussions

This strategy was assessed and a proof-of-concept is detailed in Chapter 4 (section 4.4.2 Human milk sample investigation). If the approach illustrated in Figure 2.17 appears in theory quite simple, the data processing and compounds identification steps remain prominent parts of the NTS workflows and is recognised as the bottleneck of NTS approaches. At the present time, the most reliable approach to annotate a maximum of signal is performed with already existing database. This peak annotation is complex as libraries dealing with exogenous compounds are

- 109/296 - Chapter 2. Instrumental method development and data processing tools numerous, while most of these libraries are suffering of a lack of information regarding metabolites of organic contaminants. Therefore, a common and consolidated library of organic contaminants and their metabolites appears as a pre-requisite need to annotate in a first step already known markers of exposure. This action is currently led by the HBM4EU project as detailed in the Chapter 2 (section 2.4.1.4 HBM4EU database). In a second step, all others unannotated peaks should be investigated to detect unknowns. An another and probably bigger issue remains in the annotation of these peaks to distinguish endogenous from exogenous compounds in order to further investigate exogenous compounds. This issue could be solved by analysing certified reference material of human matrix free of organic contaminants. Unfortunately, it does not exist yet and it seems very challenging as all signals should be characterised in order to offer either the best material purity without contaminants or material with known contaminant signatures which can then be managed with bioinformatics tools. The combination of both challenges, i.e. annotation of known contaminants and discrimination of unknown contaminants from endogenous compounds, make the annotation step very complex. Finally, exogenous contaminant detected in a sample extract does not only represent markers of exposure but also external contamination occurring during sample collection, storage, preparation and analysis. This issue related to the external contamination is further detailed in this chapter (section 2.5 Management of the external contamination).

Once a compound has been annotated, the last step is to confirm its identity. As detailed in the Chapter 1 (1.4.4.2 Compound annotation and identification), different confidence level according to different scale exist and at the present time there is no consensus regarding this aspect. If the compound is annotated with a library based on the MS spectrum, MS/MS experiment should be conducted to increase the level of confidence. The ideal case is to inject the analytical standard of this compounds in the same chromatographic and MS conditions. However, analytical standard is not systematically commercially available and this is especially true for unknown compounds.

 Perspectives regarding the developed strategy

The recent development of an updated version of HaloSeeker v2.0 is expected to manage multi- file alignment, to deal with procedural blank samples and to consolidate the annotation through the integration of all related adducts and fragments in the considered structural information with CAMERA tool (Kuhl et al., 2012). Then, the sensitivity appears as a second pivotal issue of

- 110/296 - Chapter 2. Instrumental method development and data processing tools these approaches. Indeed, concentration levels of exogenous contaminants in human biological matrices remain in general very low and a combination between efficient sample preparation and highly sensitive instrumentation are crucial to make the investigation possible. Finally, recently developed and updated bioinformatics software are getting powerful to distinguish background signal to compound. This also allows to reach lower concentration level.

For the present strategy, MS/MS experiments were carried out when a signal was primarily annotated with the HBM4EU list. MS/MS information is needed to validate the identification of a compound and it can also facilitate the annotation when it is queried in a library. Two advanced data acquisition approaches have been developed to make this more automated, namely the data dependent and data independent acquisition (DDA and DIA). DDA allows to fragment “n” ions passing a pre-set threshold of intensity whereas DIA provides MS/MS spectra for all ions. In the first case, only the most intense ions are fragmented. In the present case study, as the most intense ions are generally endogenous compounds remaining in the final extract, DDA does not appear as the most suitable option. On the other hand, advanced software are required to deconvolute DIA spectra. Also, exogenous compounds are at very low concentration which implies that only a few scans per peaks are detected and DIA approach could miss those because of the large dynamic range and the limitation of scan speed. A compromise between both DDA and DIA has been found based on iterative DDA. This approach has been adapted from Koelmel et al. 2017 work and recently implemented in LABERCA (June 2020) after slight modifications in the script. Following a first full-scan acquisition with top-5 DDA, an exclusion list is created to discard these top-5 most intense ions and run a second injection and then fragment next top-5 ions. After the second injection, a second list of exclusion is generated combining the previous one and last fragmented ions (Figure 2.18). Following this step-by-step approach, an extended number of ions, even at low concentration can be fragmented without a priori knowledge and with sufficient mass accuracy. The two main purposes are then i) to get MS/MS data of compounds previously prioritised with the non-targeted workflow, and ii) to be able to query MS/MS library to annotate more signals. This part of the study is still at the first stage of development and advanced software as MS- DIAL appears as a promising tool to process this data.

- 111/296 - Chapter 2. Instrumental method development and data processing tools

Figure 2.18: Iterative-DDA strategy used in LC-HRMS to generate MS/MS spectra. For the present example, a single sample is injected three time within Top-5 DDA condition. An exclusion list is created after each injection and merged to the previous exclusion list in order to fragmented less and less intense ions.

Creation of LC-HRMS database

As detailed in the previous sections, compound annotation is possible if a database including all contaminants and their metabolites with sufficient spectral information exists. An action regarding this task is in progress within the HBM4EU project. In addition to the difficulty to generate harmonised data between instruments and laboratories, it is a long task where many hours of analysis and data processing are needed. LABERCA has been involved in this action and a MS method was developed during the present Ph.D. work, not only to contribute to this task by generating MS and MS/MS reference data for a range of compounds available in the laboratory (including compounds of the QA/QC mix 2), but also to analyse analytical standard matching with a compound detected in human sample. In the first case, a short LC-HRMS method over 5 minutes was developed with full-scan acquisition and MS/MS fragmentation at normalised collision energy (NCE) of 10, 20, 30, 40, 50 and 60 (Figure 2.19). Acquiring spectra at six collision energies in a single run induces a decrease in mass resolution to get enough scan per peak in order to generate clean fragmentation spectra (method details are in

- 112/296 - Chapter 2. Instrumental method development and data processing tools

Appendix 4). In the second case, the analytical standard was analysed with the same MS method to select the best collision energy, which should be applied on the compound detected in the sample.

Figure 2.19: Examples of reference MS/MS spectra generated for fipronil (20 ng µL-1) at NCE 10, 20, 30, 40, 50 and 60 acquired in a single acquisition to input into the annotation database developed for markers of chemical exposure within the HBM4EU project.

Investigation of GC-HRMS data

The strategy to investigate GC-HRMS data is less elaborated than the strategy for LC-HRMS data, not only for timing reasons related to the recent implementation of the instrument, but also because of the uncommon nature of data and the challenging access to a library. For GC-MS data annotation, some databases already exist, such as NIST, but it is a unit based (Stettin et al., 2020; Kwiecien et al., 2015) and no database equivalent to NIST incorporate HRMS spectra.

- 113/296 - Chapter 2. Instrumental method development and data processing tools

In addition, EI ionisation generates many fragments. In general, the molecular ion either is at a very low intensity or not detected. Consequently, it is not feasible to annotate compounds based on the mass of the molecular ion, and the approach based on the HBM4EU list used to annotate LC-HRMS data is not applicable. Moreover, many molecules are fragmented at the same retention time, which makes the interpretation more complex because of spectral overlap.

We recently started the elaboration of a homemade database with spectra acquired on GC- Orbitrap. However, it is not large enough to cover thousands of compounds and, to this day, mainly includes contaminants from the QA/QC mix 2. In this context, GC data were manually processed by extraction of ion chromatograms with Xcalibur with a mass tolerance set at 5 ppm. Fortunately, encouraging solutions are emerging to efficiently process GC-HRMS data, including deconvolution tools such as embedded in MS-DIAL. This software was investigated during the collaboration with RECETOX and appeared as a very promising tool. Following the deconvolution by MS-DIAL it is possible to match spectra against the NIST library. Unfortunately, NIST spectra were mainly acquired in LRMS (unit-based resolution) and the added-value of HRMS is not usable (Stettin et al., 2020). Additionally, the GC-Orbitrap geometry influences ion ratio which in turn directly impacts the score of similarity. Another approach to annotate GC-EI-HRMS is to create another database with HRMS data. Following the collaboration with RECETOX, compounds of the QA/QC mix 2 were analysed and spectra recorded in a home-made database. A user-guide was established to facilitate this database creation (Appendix 5). As for LC data, it is a long task and as a first intention it appears more realistic to find a way to match HRMS data against LRMS data. Kwiecien et al. (2015) proposed to convert HRMS into LRMS to match fragmentation spectra with the NIST. Each match generates putative formulas that are matched against the high resolution spectra. This script was implemented in LABERCA in July 2020 and only few data have been investigated with this approach. During the present Ph.D. work and as a first suspect screening, GC-HRMS data were manually investigated by extraction of ion chromatograms with Xcalibur with a mass tolerance set at 5 ppm, with a particular focus on standards of the QA/QC mix 2 already recorded in the homemade database. However, it is not large enough to cover thousands of compounds and additional work is required to enrich the database or to be able to use NIST with the advantage of HRMS.

- 114/296 - Chapter 2. Instrumental method development and data processing tools

2.5. Management of the external contamination

The previous section 1.4.5 introduces the challenges related to the management of the external contamination. The present section deals with these issues and how they were addressed during the present Ph.D. project. LC- and GC-HRMS methods were optimised in order to screen a broad range of molecules. It also includes the detection of all external contamination occurring through different sources during sample preparation and analysis. As ubiquitous compounds are sought, they can be found in the laboratory, in material (especially in plastics), solvents and standards. For targeted methods, if the external contamination is composed of compounds of interest, the origin of the contamination has to be identified in order to control it and to ensure an accurate quantification. For non-targeted methods, the nature and sources of such external contamination are even wider because almost all injected compounds are detected. In addition, compounds from the external contamination can promote the ion suppression phenomenon. Furthermore, external contamination increases the number of detected compounds and it makes the processing of data more complicated. Although this procedural contamination cannot be realistically completely discarded or avoided, some cautions and appropriate QA/QC procedures should be considered to manage this issue.

In the context of the present work, solvent quality was of high grade (i.e. LC-MS quality) and all analytical standards were of high purity (<98%). In addition, all glassware was baked at 400 °C overnight. Pipette tips were equipped of a filter to limit cross contamination between users and applications. Fume hood was not in a positive air pressure clean room but it was systematically cleaned from top down before any experimentation. Despite this effort, residual external contamination cannot be totally discarded and has to be managed during data processing, if this contamination is repeatable. To ensure this repeatability all experiments were composed of three procedural blanks. A procedural blank was a sample without matrix or with solvent to mimic the matrix. It was treated with the same protocol as other samples and analysed in the same batch.

Then, the data processing enables to make the difference between procedural contaminations and compounds originally present in the matrix. Different approaches exist to manage this contamination. The more stringent methodology, and the one used for the present Ph.D. work, is to discard all compounds detected at least once in a procedural blank. It enables to prioritise the data processing around compounds only detected in the sample matrix. This approach is

- 115/296 - Chapter 2. Instrumental method development and data processing tools interesting as a first intention to prioritise data. However, as compounds can be present in both matrix and procedural blank samples with different concentration, the characterisation of the sample is not complete. To extend the consideration of the external contamination, a semi- quantitative approach allows to better characterise a sample. It is based on the subtraction of signal detected in blank to the matrix sample, but good repeatability of the external contamination is required. Such of these options are available in HaloSeeker v2.0 and the signal is displayed if it is a certain number of times more intense in the sample than in the blank. The reference signal in blank can be the maximum detected intensity or the average of intensities detected in all blanks. At the present time no official guidelines have been issued to manage the external contamination with non-targeted screenings and the existence of different approaches illustrates the complexity of the task.

2.6. Conclusion

This second chapter is dedicated to the presentation of instrumental methods and data processing strategies developed during the present Ph.D. work in order to perform non-targeted analyses from human samples. It highlights the compromises that have to be made to detect a large range of molecules. The complementarity of LC- and GC-HRMS was demonstrated with the QA/QC mix 2, elaborated for guiding this development and assessing method performance. For both instrumentations, generic settings were selected in order to analyse molecules with various physicochemical properties. As human samples treated with non-targeted screening approaches lead to complex extracts containing matrix interferences, reinforced preventive maintenance provisions were defined as necessary. They were implemented within the analytical workflow in order to ensure good analytical reliability over the batch and compatibility with the data processing requirements. This last part was also largely investigated and many interesting perspectives have been drawn. LC- and GC-HRMS data were separately investigated with different strategies based on the nature of the data. For LC, a strategy combining the recently developed software HaloSeeker with a manual investigation step was developed to highlight signals of interest. The bottleneck remains the compound identification. This make the prioritisation of unknown compounds more complex because even known compounds are not annotated. On one hand, iterative DDA approach is promising to acquire MS/MS spectra comparable with already existing databases. On the other hand, HBM4EU partners are working on a well referenced and common database for exposomics researches. Regarding GC, MS and MS/MS information are already present in the spectra as EI is a highly

- 116/296 - Chapter 2. Instrumental method development and data processing tools energetic ionisation source. This prompts to investigate the whole spectra and not only the molecular ion, which sometimes is not detected because of high fragmentation. Then, the NIST library already gathers thousands of spectra, but most of them are at unit resolution and exact mass is not exploitable although it can be of high interest to facilitate the identification. Recent approaches involving bioinformatics tools are in progress to combine both and on the short-term appear more efficient than building again a new library. In the meantime, the strategy used during the present Ph.D. work was based on manual approach by extracting ion chromatogram obtained for already known compounds.

The present LC- and GC- analytical methods with dedicated data processing can be used to characterise real samples, keeping in mind the current limitations and perspectives. The investigation of real samples also stimulates both the development and improvement of analytical methods and data processing. Thus, other limitations can be highlighted for instance the sensitivity depending on the matrix and the compatibility of the data processing with acquired signals. Different matrices were analysed with various sample preparation and are detailed in Chapter 3.

- 117/296 -

3. CHAPTER 3

DEVELOPMENT OF NON-TARGETED SAMPLE PREPARATION PROTOCOLS TO SCREEN HALOGENATED MARKERS OF CHEMICAL EXPOSURE IN HUMAN SAMPLES

- 119/296 -

Chapter 3. Development of non-targeted sample preparation protocols to screen halogenated markers of chemical exposure in human samples

3.1. Introduction

Non-targeted screening approaches appear today as a good option to characterise the human chemical exposome. However, no guidelines have been yet proposed to perform NTS from human matrices, even though this particular application implies specific and significant analytical challenges, for which such concerted and rationalised guidelines now appear necessary (method assessment, identification, procedural blank management). While of already significant and growing maturity in the environment and food safety areas respectively, this large-scale screening issue remains poorly addressed in the human biomonitoring field. Possible explanations for this observation are (i) the lower availability of samples and in lower quantity, (ii) the lower expected chemical concentration levels and /or higher complexity of human matrices compared to some environmental and food matrices, meaning an additional level of analytical difficulty, (iii) the occasional presence of hitherto unidentified biotransformation products (metabolites) in human matrices, rather than pre-identified parent compounds, and (iv) the lack of structuration of the necessary HBM networks and laboratories compared to other fields.

As mentioned in Chapter 1 (section 1.4 Challenges of non-targeted screening workflows), one of the main issues to be considered during the development of NTS method is the definition of an appropriate level of selectivity of the selected protocol. Indeed, the LC- and GC-HRMS systems used for NTS are not compatible with direct injection of complex liquid or solid matrices. Even when the final extract appears compatible with the system of injection, matrix interferences remaining in the final extract could impact the detected signal through ion suppression phenomenon and other matrix effect related troubles. Therefore, a preliminary extraction followed by a certain level of purification is mandatory. On the other hand, in non- targeted approaches, the widest possible range of compounds should be detected and thus the extraction and purification should be as exhaustive and non-selective as possible. This is more challenging for human matrices because low volumes are generally available, so the starting amount should be low but compatible with the sensitivity of the protocol. Accordingly, the number of possible experiments of development is also reduced. Finally, one of the main challenges associated with NTS is to find a good balance regarding purification selectivity in order to limit matrix interferences while preserving as many compounds of interest as possible. Whatever the sample preparation technique applied, these issues have to be assessed using a

- 121/296 - Chapter 3. Development of non-targeted sample preparation protocols to screen halogenated markers of chemical exposure in human samples mix of appropriate reference standards, ideally representative of the desired application scope of the method.

At the present time, no unique sample preparation method is capable to comprehensively cover a large range of markers of exposure with diverse properties. As illustrated in Figure 3.1, plotting logKow as a function of the monoisotopic mass of known emerging chemicals, the majority of compounds (~90%) have logKow-values within -5 to 10 and monoisotopic mass within 100-1000.

Figure 3.1: logKow-values as a function of monoisotopic masses for compounds included in the Merged NORMAN Suspect List inventorying known emerging chemicals in environmental matrices (SusDat, 2012).

In this context, the aims of the present chapter are the evaluation of the crucial methodological issues related to the sample preparation and the assessment of several protocols in order to develop a NTS strategy applied to various human matrices. The purpose is to highlight the necessary compromises and/or to propose guidelines in order to implement NTS approach in both laboratories wishing to join this dynamic and biomonitoring studies (i.e. analysis of real samples).

In the frame of the HBM4EU project, both liquid (human milk) and solid (placenta, meconium, adipose tissue) matrices were considered to develop such NTS approaches, with a special focus on the characterisation of the chemical exposure during the perinatal period. In addition to this physical distinction, the lipid content of the considered matrix also strongly impacts the sample

- 122/296 - Chapter 3. Development of non-targeted sample preparation protocols to screen halogenated markers of chemical exposure in human samples preparation process to be applied. A large range of organic contaminants have lipophilic properties (e.g. POPs) so they are extracted with lipids and the separation of these contaminants from lipids is challenging. Moreover, lipids should be removed because they are detected in LC- and GC-MS and their signal would make the data processing more complex. Also, it can promote the ion suppression phenomenon, leading to the signal suppression of contaminants of interest. In order to overcome the physical diversity and investigate a large range of compounds, several options were envisaged in terms of sample preparation strategies, from widely used liquid-liquid extraction (Bligh and Dyer method) to methods that are selective towards lipids, such as acidic hydrolysis, solid phase extraction on Captiva EMR-Lipid® stationary phase, or fractionation through GPC or EDA approach.

In the present chapter, the different sample preparation strategies tested and finally selected for application of NTS to various human matrices are described. Solutions to reduce the risk of external contamination, including the interpretation of procedural blank samples, are also presented. In addition, a methodology was proposed to assess the method performance, leading to the creation of a predictive tool highlighting method limitations.

3.2. Acidic hydrolysis and GPC: application to adipose tissue

3.2.1. Introduction

The development of a non-targeted sample preparation was previously initiated in LABERCA on biota samples (chemical exposure characterisation in sentinel species, for instance eel muscle, also containing a high amount of lipids). A dedicated protocol was developed for such environmental matrices, including an acidic hydrolysis to digest and remove fat content (Cariou et al., 2016). This step is based on a liquid-liquid partitioning between an organic phase to solubilise lipophilic compounds and a very acidic aqueous phase to hydrolyse lipids. As human adipose tissue contains approximately 90 % of lipids, the same method was first considered in the context of our own study. This approach was then compared to a less aggressive method based on molecules separation according to their size, the gel permeation chromatography (GPC). With the last technical option, lipids are partly removed since their size (> 600 Da) exceeds the size of most of organic contaminants (< 700 Da).

- 123/296 - Chapter 3. Development of non-targeted sample preparation protocols to screen halogenated markers of chemical exposure in human samples

3.2.2. Acidic hydrolysis protocol

Sample preparation and analysis

All pieces of glassware were baked at 400 °C overnight to minimise external contamination. Lipids and lipophilic compounds were extracted from 5 g of fresh adipose tissue by pressurised liquid extraction (PLE) using a Speed Extractor E-914 (Büchi). PLE cell was filled with approximately 1 g of celite and the sample on top. Sample was extracted thanks to three successive static cycles with a mixture of toluene/acetone (70:30, v/v) at 120 °C and 100 bar. The organic extract was evaporated until a 1 mL volume with a rotary evaporator and until dryness at room temperature and atmospheric pressure during two days. The extracted fat (4.51 g, 88% fresh weight) was suspended into hexane (27 mL). One gram of extracted fat fortified with a mixture of 13C-TBBPA and 2H-γ-HBCDD (4 ng each) was treated by successive liquid- liquid partitioning with concentrated sulphuric acid (4 x 4 mL) to remove lipids. The organic layers were neutralised with 10 mL ultrapure water and dried with anhydrous sodium sulphate. The purified extract was spiked with 4 ng of 2H-α-HBCDD in toluene, evaporated to dryness under a gentle stream of nitrogen and reconstituted in 40 µL of acetonitrile/water (80:20, v/v). Samples were analysed by LC-HRMS (method details in Chapter 2, section 2.2.2 LC-HRMS method development), with the elution gradient G1, in negative ionisation mode. For all the experiments presented in this manuscript, internal and external standards were labelled (2H or 13C) to distinguish spiked standards from contaminants present in the sample.

Matrix effect

A human adipose tissue sample and a procedural blank (both unspiked) were prepared and analysed according to the protocol detailed above. Total ion currents obtained for both samples are represented in Figure 3.2. The signal detected in adipose tissue remains concentrated after the purification. This density of signals could impair the detection of analytes by matrix effect. In order to better characterise the efficiency of the sample preparation to remove matrix interferences, the matrix effect was assessed.

- 124/296 - Chapter 3. Development of non-targeted sample preparation protocols to screen halogenated markers of chemical exposure in human samples

Figure 3.2: Overlap of total ion currents obtained for a human adipose tissue (blue) and a procedural blank (black) samples purified with an acidic hydrolysis and analysed by LC-ESI(-)-HRMS.

Matrix effect and more precisely ion suppression is a phenomenon occurring in the ion source and can impact the signal intensity. Ion suppression can happen when co-eluted compounds (e.g. matrix interferences, external contaminations, multiple CEC) are more easily ionised than targeted compounds and can lead to the non-detection of these analytes. Ionisation enhancement can also occur in the source when co-eluting compounds promote the CEC ionisation. For non-targeted method, the matrix effect is difficult to characterise because it should be known for all compounds, at all RT, which is not feasible. However, we have studied this matrix effect on a sub-set of standards. For the present sample preparation of adipose tissue by acidic hydrolysis, a mix of analytical standard was used to evaluate the ion suppression phenomenon. A solution of hydroxy-trichlorodiphenyl ether (HtCDE), hydroxy- tribromodiphenyl ether (HtBDE) and tetrabromobisphenol A (TBBPA) at 1 ng µL-1 was prepared in acetonitrile. In parallel of a sample injection, the solution of standards was post-column infused at a flow rate of 5 µL min-1 (Figure 3.3). The resulting total ion currents (TIC) is reflecting the global stability of the overall signal arriving to the detector (sum of all ionic species from the standard compound and the sample extract). The specific extracted ion of the standard compound (EIC) can be affected in some part of the chromatogram by interferences from the matrix that leads to a decrease of this signal.

- 125/296 - Chapter 3. Development of non-targeted sample preparation protocols to screen halogenated markers of chemical exposure in human samples

Figure 3.3: Illustration of the typical experiment set up to characterise the ion suppression phenomenon by LC-HRMS. A pure standard of a given compound is post-column introduced in the detection system together with the prepared sample extract. The fluctuation on the extraction ion chromatogram characterise matrix effect. Adapted from Antignac et al., 2005.

The matrix effect assessment was carried out for both adipose tissue and procedural blank samples (Figure 3.4) for the three standards (HtCDE, HtBDE and TBBPA). For both HtCDE and HtBDE the signal was almost completely suppressed between 9.5 and 11.5 min as it almost reached an intensity of zero. This demonstrated the negative effect of ion suppression and how it can drastically compromise the detection of several compounds. However, in the present case, if the three standards are present in a sample they will not be affected by the ion suppression because the phenomenon occurred approximately one minute after their elution.

- 126/296 - Chapter 3. Development of non-targeted sample preparation protocols to screen halogenated markers of chemical exposure in human samples

Figure 3.4: Overlap of the LC-ESI(-)-HRMS extracted ion chromatograms of hydroxy- trichlorodiphenyl ether (HtCDE), hydroxy-tribromodiphenyl ether (HtBDE) and tetrabromo bisphenol A (TBBPA) observed for typical adipose tissue (blue), procedural blank (red) and mobile phase (black) samples (from top to bottom) in presence of a post-column introduction of a solution of HtCDE, HtBDE and TBBPA. Last chromatograms (bottom) show the observed signals for the three reference compounds injected as the mix of standards without post-column infusion.

This qualitative experiment allows to visualise the matrix effect phenomenon. In order to quantify the signal loss or gain, the signal of reference standards detected in a sample fortified after the application of the sample preparation procedure should be compared to the signals of the same reference standard injected in pure solvent at the same concentration level. The difference of intensity characterises the matrix effect by signal enhancement or suppression. An example of such assessment is available in a further the section of this manuscript, dedicated to human milk (3.4.3.4 Matrix effect). This experiment only takes into account the matrix effect

- 127/296 - Chapter 3. Development of non-targeted sample preparation protocols to screen halogenated markers of chemical exposure in human samples during the ionisation. It can be conducted for each sample preparation and each matrix to better characterise the efficiency of the treatment.

3.2.3. GPC protocol

Sample preparation

Prior to this investigation, the glass equipment was burned at 400 °C overnight to minimise external contamination. Lipids were extracted with the same PLE method detailed in the previous section. One gram of extracted fat was divided in 5 vials containing 0.2g, each dissolved in 500 µL of ethyl acetate/cyclohexane (1:1 v/v) and purified by GPC. The injection was performed with a flow rate of 5 mL min-1 ethyl acetate/cyclohexane (1:1 v/v) over 70 min, in isocratic mode. Extract was collected between 20 and 70 min (see development below), resulting in 250 mL of solvent/sample concentrated with rotary evaporator until 1 mL and transferred in a vial. Purified extract was spiked with 4 ng of 2H-α-HBCDD in toluene, evaporated to dryness under a gentle stream of nitrogen and finally reconstituted in 40 µL of acetonitrile/water (80:20, v/v).

Instrumental method

The GPC column (580 x 24.4 mm internal diameter glass column) was manually packed with Bio-Beads SX-3 (40 cm height in the column) swelled in acetate/cyclohexane 1:1 (v/v). Instrument performance was checked with a mix of sunflower oil, bis(2-ethylhexyl) phthalate and perylene in ethyl acetate/cyclohexane 1:1 (v/v). The wavelength of detection with ultraviolet detector was set at 254 nm. The method of elution was previously developed according to the EPA 3640A method with ethyl acetate/cyclohexane 1:1 (v/v) instead of dichloromethane (Bichon et al., 2015). The GPC collection windows was evaluated by injecting a sample of adipose tissue in order to detect the elution time of fat. The extract was fractionated by 2 min segments from 0 to 70 min and collected in previously weighed tubes. Each fraction was evaporated until dryness under a gentle stream of nitrogen and tubes were weighted to calculate the amount of fat as a function of the collection window (Figure 3.5-A). In agreement with this last result, UV detection shows peaks of adipose tissue fat content from 10 to 20 min (Figure 3.5-B) and this window should not be collected so as to remove lipids from samples. Furthermore, the Cl/Br-phenolic compounds constituting the standard mix 1 were injected (4 ng in 500 µL of ethyl acetate/cyclohexane 1:1 (v/v)) and fractionated by 2 min segments from

- 128/296 - Chapter 3. Development of non-targeted sample preparation protocols to screen halogenated markers of chemical exposure in human samples

0 to 70 min to determine the time of elution. Each fraction was analysed by LC-ESI(-)-HRMS to determine the fraction of elution. The results provided in Figure 3.5-C indicate that standards were eluted by GPC between 20 and 50 min. This confirms the choice of a collection window between 20 and 70 min.

- 129/296 - Chapter 3. Development of non-targeted sample preparation protocols to screen halogenated markers of chemical exposure in human samples

Figure 3.5: Optimisation of the GPC fractionation purification. Gravimetric determination (A) and UV signal at 254 nm (B) of the fat content of adipose tissue fractionated by 2 min segments. LC-ESI(-)-HRMS analysis of fractions collected by GPC from a sample of Cl/Br-phenolic compounds of standard mix 1, fractionated by 2 min segments. Extracted ion

- 130/296 - Chapter 3. Development of non-targeted sample preparation protocols to screen halogenated markers of chemical exposure in human samples

chromatograms for each standard in each fraction were integrated and the peak area is reported in C.

Comparison of GPC with acidic hydrolysis

Adipose tissue purification by acidic hydrolysis is an efficient approach to remove lipids. However, it is based on a liquid-liquid extraction where the polar fraction cannot directly be analysed because of very acidic pH and residues of fat while this polar fraction could contain compounds of interest. The GPC was then envisaged as an alternative option to better preserve the integrity and comprehensiveness of the analysed sample. The same adipose tissue as previously purified with the acidic hydrolysis approach was prepared with the GPC (matrix and blank samples spiked with the labelled standards only) and analysed in the same LC-HRMS conditions. The Figure 3.6 shows the resulting total ion currents obtained for both adipose tissue and procedural blank purified by GPC. As for the acidic hydrolysis method, the obtained fat sample extract was found to present significant matrix interferences. Furthermore, a lot of peaks were also detected in the procedural blank. This external contamination observed with GPC was estimated to drastically compromise the further interpretation and identification, since the compounds detected in a sample could come from the procedure and could promote ion suppression. This external contamination issue appears as a main concern with regard to non-targeted approaches and consequently the present case was more deeply investigated, as detailed in the next section.

Figure 3.6: Overlap of LC-ESI(-)-HRMS total ion currents obtained for adipose tissue (blue) and procedural blank (black) samples purified according to the developed GPC procedure.

- 131/296 - Chapter 3. Development of non-targeted sample preparation protocols to screen halogenated markers of chemical exposure in human samples

In addition to the significant external contamination issue, the GPC approach used for our particular application was characterised by other drawbacks. In comparison with the acidic hydrolysis purification, able to treat 1 g of fat in ~2h with approximately 10 mL of sulphuric acidic and 5 mL of hexane, the used GPC column is indeed limited at 200 mg of fat per injection. Thus, five GPC injections are needed to prepare 1 g of fat, requiring a total of 1250 mL of solvent and 350 min of run, plus the time required to evaporate this large volume of solvent. Then, the GPC approach appeared significantly longer and as a less “green” technique than the acidic hydrolysis. According to these issues, and in the current state of the development and optimisation, the purification by GPC was not further retained for our particular application. However, in the global context of non-targeted analysis, the purification by fractionation theoretically remains more suitable than the acidic hydrolysis. It should help to preserve the integrity and comprehensiveness of the accessible markers, and complementary purification protocols could be conducted on specific fractions of higher interest. Even if this fractionation approach was not further investigated during the present Ph.D. project, it deserves to be deeply considered and optimised in future works with additional allocated time and efforts.

Investigation of external contamination by GPC

The procedural blank sample previously prepared through GPC fractionation and analysed by LC-HRMS (gradient G1) was investigated with HaloSeeker for more comprehensive data processing. TBBPA was rapidly confirmed to be present in this procedural blank (Figure 3.7). Other clusters were revealed in the same procedural blank extract but not identified, considering the time required for such task and the fact that GPC was not selected for further investigation. However, the particular case of TBBPA external contamination was further investigated to identify its main sources.

- 132/296 - Chapter 3. Development of non-targeted sample preparation protocols to screen halogenated markers of chemical exposure in human samples

Figure 3.7: MD plot obtained with HaloSeeker v1.0 software following the data processing of a procedural blank sample purified by GPC and analysed by LC-ESI(-)-HRMS (top). The particular signal of TBBPA is indicated in the black circle. Experimentally observed mass spectra (bottom left, down red traces) and extracted ion chromatogram (bottom right) corresponding to this signal.

Consequently, different possible sources of external contamination were considered for TBBPA. At first glance, the three main sources of potential external contaminations were the GPC system, the fume hood used for samples collection and the rotary device used for extract

- 133/296 - Chapter 3. Development of non-targeted sample preparation protocols to screen halogenated markers of chemical exposure in human samples evaporation. To evaluate these hypotheses, specific samples were prepared for analysis, according to the following protocols:

- Contamination from the GPC system: six vials containing 500 µL of ethyl acetate/cyclohexane 1:1 (v/v) and 4 ng of 2H-γ-HBCDD and 13C-TBBPA were injected onto the GPC system and each 2 min fraction was collected between 20 and 70 min of elution. Solvent was evaporated until 1 mL with rotary evaporator previously cleaned with dichloromethane. Then it was transferred in vial and evaporated under a gentle stream of nitrogen until dryness and resuspended in 40 µL acetonitrile/water 8:2 (v/v) containing 4 ng of 2H-α-HBCDD. - Contamination from the fume hood: a round-bottom flask filled with 250 mL of ethyl acetate/cyclohexane 1:1 (v/v) and 4 ng of 2H-γ-HBCDD and 13C-TBBPA was left overnight under the fume hood without lid. This volume corresponds to the volume collected per sample between 20 and 70 min of elution. Solvent was evaporated with the same protocol (rotary evaporator and under a gentle stream of nitrogen) as the samples prepared to assess the contamination through the GPC system. - Contamination from rotary evaporator: three round-bottom flasks filled with 125; 250 and 500 mL of ethyl acetate/cyclohexane 1:1 (v/v) and 4 ng of 2H-γ-HBCDD and 13C- TBBPA were evaporated until 1 mL on the same rotary evaporator previously cleaned with dichloromethane. Final extract was prepared with the same protocol as the samples prepared to assess the contamination through GPC system.

The applied sample treatments and instrumental stability were validated between all samples with internal (2H-γ-HBCDD and 13C-TBBPA) and external (2H-α-HBCDD) standards that exhibited signal variability (RSD %) of 8%, 21% and 14%, respectively. The comparison of total ion currents (Figure 3.8) obtained for one procedural blank extract prepared by GPC, two procedural blanks prepared without GPC (V = 250 mL and blank left overnight under the fume hood) illustrated the significant contamination from the GPC system. More specifically, TBBPA was only detected in samples prepared with GPC, with 55 % signal variability. The GPC system was then identified as the main source of external contamination to TBBPA. The observed variability also demonstrated a need for a deeper assessment and management of this phenomenon. In any case, the systematic inclusion of several procedural blanks during non-targeted screening experiments is crucial to avoid false positive results.

- 134/296 - Chapter 3. Development of non-targeted sample preparation protocols to screen halogenated markers of chemical exposure in human samples

Figure 3.8: Overlapped LC-ESI(-)-HRMS total ion currents (top) observed for procedural blank samples prepared with GPC (blue), rotary evaporator (red), left overnight under the fume hood (green) and mobile phase (black). Extracted ion chromatogram of TBBPA observed in procedural blank sample prepared with GPC (bottom left) and the corresponding mass spectrum annotated with mass deviation in brackets (bottom right).

This study of external contamination is not exhaustive and the detection of TBBPA was given as a concrete example to illustrate the importance to include procedural blanks in all NTS experiments. The full identification and characterisation of the different sources of contamination is already a vast task for a given known compound monitored by conventional targeted approaches. It is worse with non-targeted screenings, due to the possible extent of involved substances. Considering the complexity of NTS data, this issue should be managed with bioinformatics tools during the data processing. At the present time, the two main procedures possibly used during the data processing with regard to those signals present in

- 135/296 - Chapter 3. Development of non-targeted sample preparation protocols to screen halogenated markers of chemical exposure in human samples procedural blank samples are i) signal subtraction (this implies that the contamination has to be repeatable) and ii) signal exclusion (signals are not literally suppressed, but not investigated). The first approach enables to investigate compounds present in both matrix and blank samples. However, it may be penalised in case of substances present in biological samples at concentration lower than the concentration present in blank. The second approach is more drastic and could limit the number of false-positive results. At the present time, no guidelines have been established and users should report the selected method.

3.2.4. Discussion and conclusion

This third chapter began with the analysis of very complex matrices due to their high lipid content. Beyond this complexity of adipose tissue, this biological compartment remains an interesting matrix to seek lipophilic contaminants. In the frame of the present Ph.D. work, two sample preparations were tested to eliminate lipids either by degradation (acidic hydrolysis) or size exclusion (GPC). Strengths and limitations of both approaches were evaluated on the basis of identical adipose tissue sample prepared according to both procedures.

Acidic hydrolysis was deemed a good sample preparation approach to remove lipids from matrix with high lipid content. It is compatible with a large volume of matrix (approx. 1 g) to concentrate compounds of interest. However, very polar compounds are less recovered with this approach than polar to nonpolar ones. Nevertheless, this limitation is acceptable to prepare adipose tissue because most of lipophilic compounds are generally polar to nonpolar. In parallel, GPC appeared as a satisfying approach to separate lipids from other contaminants. However, the present study highlighted a number of limitation for GPC in terms of solvent consumption, time of analysis and external contamination. However, this fractionation approach is, in theory, in adequacy with non-targeted method development, combined with other purification step. For instance, fractions without lipids could be directly injected rather than fractions with lipids could be further purified by acidic hydrolysis. This way, a wider range of compounds could be potentially recovered. According to these conclusions, neither acidic hydrolysis nor GPC were used to treat other matrices with lower lipid content than adipose tissue. Alternative sample preparations based on partitioning were investigated and are presented in the following sections.

- 136/296 - Chapter 3. Development of non-targeted sample preparation protocols to screen halogenated markers of chemical exposure in human samples

3.3. Bligh and Dyer method: application on placenta

3.3.1. Introduction

Bligh and Dyer proposed in 1959 an eponym protocol for lipidomics studies from fish muscle containing approximately 1% of lipids. This approach, as similar others (Folch et al., 1957; Matyash et al., 2008), can be used for metabolomic and lipidomic studies of blood (Patterson et al., 2015). Based on a single liquid-liquid extraction, this approach is also interesting for NTS as all compounds are preserved and samples can be comprehensively analysed. Polar to very- polar compounds as most of metabolites are expected in the polar (water/methanol) fraction whereas lipids and other polar to nonpolar compounds are expected in the nonpolar fraction (chloroform). Moreover, it was designed to analyse reduced amount of samples which is also an important element to keep in mind for the development of NTS method on human matrices. For the present Ph.D. project, dealing with the characterisation of chemical exposure, especially at early stage of life, the Bligh and Dyer method was adapted to analyse placenta as this matrix contain approximately 1 % of lipids.

3.3.2. Method development

Protocol

Grinded placenta (100 mg) was weighed in Precellys® tube (2 mL) containing ceramic beads. Procedural blank was considered as an empty Precellys® tube (2 mL). Successively, 150 µL of water, 250 µL of methanol containing internal standards and 400 µL of chloroform were added, with 30 sec of vortex after each addition. Samples were homogenised twice during 30 sec at 6 000 rpm with a pause of 45 sec between each with a Precellys® homogeniser. After 15 min of centrifugation at 4°C and 13 000 g, denatured proteins were precipitated and suspended at the interface between polar (water/methanol) and nonpolar (chloroform) layers. 50 µL of each fraction were concentrated in a gentle stream of nitrogen until dryness. Extracts were reconstituted in 50 µL of water/acetonitrile 8:2 (v/v) and acetonitrile/water 8:2 (v/v) for the polar (water/methanol) and the nonpolar (chloroform) fraction, respectively (Figure 3.9).

- 137/296 - Chapter 3. Development of non-targeted sample preparation protocols to screen halogenated markers of chemical exposure in human samples

Figure 3.9: Bligh and Dyer protocol applied to 100 mg placenta samples assisted by Precellys® homogeniser device.

Method performance

 HRMS scan range

The Bligh and Dyer protocol allows to split compounds in two fractions regarding compounds polarity. Polar to very polar compounds, as most of metabolites, are expected in the water/methanol fraction whereas less polar to nonpolar compounds, e.g. lipids, are expected in the chloroform fraction. In order to assess the capability and performance of this approach, three unspiked placenta samples were prepared with the protocol described in the previous section and analysed by LC-HRMS, in positive and negative mode and with two scan ranges 100-1 000 Da and 1 000-2 000 Da. Total ion currents were compared (Figure 3.10) and the fraction containing the most information with sufficient retention was the chloroform fraction with negative ionisation and the 100-1 000 Da range. Compounds in the water/methanol fraction were mainly eluted at the beginning of the chromatogram and even in the dead volume. Since polar to very-polar compounds are expected in this fraction, another chromatographic system with better retention of these analytes is required. For instance, HILIC conditions can be considered fit for purpose, as it retains polar compounds and is compatible with MS detection.

- 138/296 - Chapter 3. Development of non-targeted sample preparation protocols to screen halogenated markers of chemical exposure in human samples

Figure 3.10: Total ion currents obtained for placenta samples extracted with the Bligh and Dyer approach resulting in nonpolar (chloroform, on top) and polar (water/methanol, on bottom) fractions. Extracts were analysed by LC-ESI(-)-HRMS on the scan range 100-1 000 DA (black) and 1 000-2 000 Da (red) and by LC-ESI(+)-HRMS on the scan range 100-1 000 DA (blue) and 1 000-2 000 Da (green).

- 139/296 - Chapter 3. Development of non-targeted sample preparation protocols to screen halogenated markers of chemical exposure in human samples

 Liquid-liquid partitioning

The liquid-liquid partitioning was assessed for Cl/Br-phenolic compounds of standard mix 1, selected as markers. Six placentas were analysed, among which three were not spiked and used as control and three others were spiked with Cl/Br-phenolic compounds of the standard mix 1 (4 ng/sample). Three unspiked procedural blank samples were also prepared with the same protocol. All placenta and procedural blank samples were prepared according to the Bligh and Dyer method detailed in the previous section.

None of the reference Cl/Br-phenolic compounds were detected in unspiked placentas, so the samples did not contain these contaminants and no external contamination by these compounds occurred during the sample preparation. To evaluate the partitioning of Cl/Br-phenolic compounds of the standard mix 1 between the two fractions resulting from this procedure, it was assumed that the sum of standards detected in both fractions was equal to 100 %. The main purpose was to determine which fraction should be prioritised and the recovery assessment allowed to estimate the loss during the procedure. The partitioning of compounds between the polar and nonpolar fraction was assessed, based on the ratio of the signal of standards detected in one fraction out of the sum of signal of standards detected in both fractions. As illustrated in Figure 3.11, the test reference standards were mainly detected in the chloroform fraction, except pentachlorophenol and pentabromophenol which were detected approximately at 50 % in both fractions. This partitioning is coherent as the nonpolar fraction is normally analysed for lipidomics studies, and organic compounds of interest are known for their lipophilic properties. Thus, this last fraction was prioritised to screen organic contaminants in placenta.

- 140/296 - Chapter 3. Development of non-targeted sample preparation protocols to screen halogenated markers of chemical exposure in human samples

Figure 3.11: Results of partitioning (LC-ESI(-)-HRMS detection) observed for the different Cl/Br-phenolic compounds of the standard mix 1 selected as markers, spiked in a human placenta sample, between the nonpolar (chloroform in red) and the polar (methanol/water in blue) fraction after the Bligh and Dyer protocol.

 Method recovery

The recovery of Cl/Br-phenolic compounds of the standard mix 1 was calculated for the same placenta samples as in the previous section. Recovery was the sum of the peak area detected in both polar and nonpolar fraction and compared to the signal of standards in pure solvent. As illustrated in Figure 3.12, most of recoveries were found between 20 and 70 %, except for mono-chlorophenol, di-chlorophenol and monobromophenol for which recoveries were higher than 100 %. These last compounds then appeared better detected in placenta than in pure solvent. Since they were not detected in unspiked placenta, this phenomenon is not an external contamination or, alternatively, these compounds were not initially in the matrix. This could be the signature of a signal enhancement by matrix effect. However, the experiment of ion suppression was not conducted on these compounds so this hypothesis was not verified. This high recovery would be an issue in a context of quantification. As it is not the purpose of non-targeted screening, those high recoveries were not further investigated in the context of the present work. Finally, the method is efficient to analyse all Cl/Br-phenolic compounds of the standard mix 1 with good recovery and the chloroform fraction should be analysed in priority to seek organic contaminants.

- 141/296 - Chapter 3. Development of non-targeted sample preparation protocols to screen halogenated markers of chemical exposure in human samples

Figure 3.12: Estimated recovery for the different Cl/Br-phenolic compounds of the standard mix 1, tested as markers spiked in placenta prepared with the Bligh and Dyer protocol.

Comparison with targeted method

The main difference between targeted and non-targeted screenings is the selectivity of the method which also induces difference of sensitivity. Targeted methods typically allow to reach lower LODs than non-targeted approaches. The characterisation of samples by targeted approaches allows to determine the level of contamination that should be reached with NTS method to detect the same compounds. This comparison is also useful to know if a compound is not detected by NTS because it is not included in the sample or because the analytical method is insufficiently efficient to detect it. The limit of detection depends on both sample preparation selectivity and instrumental LOD. As detailed in Chapter 2, the instrumental LOD was calculated with the dilution curve of standards from 0.001 to 0.5 ng loaded on the column. It was concluded that in the best case scenario, the instrumental LOD is 0.001 ng µL-1 or 1 pg µL- 1. From this value, it is possible to back calculate the minimal contamination level that could be detected in a sample. This concentration was calculated in the ideal case, meaning without loss during either sample preparation or analysis. If the instrumental LOD is 1 pg µL-1, it means the quantity in a vial containing 50 µL of solvent is 50 pg (1 pg/µL = 50 pg/50µL). Thus, at the beginning the sample should contain 50 pg of compound of interest. In the present case, where 100 mg of placenta were analysed and since there is 1 % of fat in placenta, the initial contamination level should be at least 50 ng g-1 fat (500 pg/g wet weight). As a comparison, Nanes et al., 2014 found maximum concentration of PCB and PBDE in placenta samples in the

- 142/296 - Chapter 3. Development of non-targeted sample preparation protocols to screen halogenated markers of chemical exposure in human samples

US at 100 and 1 000 pg g-1 ww, respectively. Thus, the present non-targeted approach could be compatible to detect contaminants in placenta samples from highly exposed women.

 Experimental protocol

In practice, six samples were analysed with a targeted method focusing on novel flame retardant (nBFR), previously developed by the department of organic contaminants of LABERCA and routinely used for food matrices (Bichon et al., 2018). Targeted compounds are listed in Table 3.1. To summarise, approximately 12 g of fresh samples were lyophilised, resulting in 2 g of dry matter which were extracted by PLE. The extracted fat (1 % of fresh placenta) was purified on several silica columns at different pH (neutral, 22 % and 44 % of sulphuric acid), and by liquid-liquid extractions. In total, four fractions were collected during the protocol and analysed with four different instruments combining LC/GC and LRMS/HRMS.

Table 3.1: List of nBFR compounds detectable with the targeted method used to characterise placenta samples.

Name Acronym Instrument 2,4,4'-tribromodiphenyl ether PBDE-28 2,2',4,4'-Tetrabromodiphenyl ether PBDE-47 2,2',4,4',5-Pentabromodiphenyl ether PBDE-99 2,2',4,4',6-pentabromodiphenyl ether PBDE-100 2,2',4,4',5,5'-Hexabromodiphenyl ether PBDE-153 2,2',4,4',5,6'-Hexabromodiphenyl ether PBDE-154 2,2',3,4,4',5',6-heptabromodiphenyl ether PBDE-183 2,2',3,3',4,4',5,5',6,6'-decabrominated diphenyl ether PBDE-209 2,2',5,5'-Tetrabromobiphenyl PBB-52 GC-HRMS 2,2',4,5,5'- Pentabromobiphenyl PBB-101 (EI-magnetic sector) 2,2',4,4',5,5'-Hexabromobiphenyl PBB-153 1,2,4,5-Tetrabromo-3,6-dimethylbenzene pTBX 2,3,4,5-tetrabromo-6-chloromethylbenzene TBCT 1,2,3,4,5-Pentabromobenzene PBBz 1,2,3,4,5,6-Hexabromobenzene HBBz Pentabromotoluene PBT Pentabromoethylbenzene PBEB Octabromotrimethylphenyl indane OBIND Decabromodiphenyl ethane DBDPE Tetrabromobisphenol A TBBPA Tetrabromobisphenol S TBBPS 4-Monobromophenol Mono-BP LC-HRMS 2,4-Dibromophenol di-BP (ESI-Orbitrap) 2,4,6-Tribromophenol Tri-BP 2,3,4,5-Tetrabromophenol Tetra-BP Pentabromophenol Penta-BP 2-Ethylhexyl 2,3,4,5-tetrabromobenzoate EHTBB APGC-MS/MS

- 143/296 - Chapter 3. Development of non-targeted sample preparation protocols to screen halogenated markers of chemical exposure in human samples

Bis(2-ethylhexyl) tetrabromophthalate BEHTBP (APCI-QqQ) Tetrabromobisphenol A dimethyl ether TBBPA bME 1,2-Bis(2,4,6-tribromophenoxy)ethane BTBPE Tetrabromobisphenol A bis(2,3-dibromopropyl) ether TBBPA-bDiBPrE Alpha-1,2,5,6,9,10-Hexabromocyclododecane α-HBCDD Beta-1,2,5,6,9,10-Hexabromocyclododecane β-HBCDD LC-MS/MS Gamma-1,2,5,6,9,10-Hexabromocyclododecane γ-HBCDD (ESI-QqQ) Tris(2,3-dibromopropyl)isocyanurate T23BPIC

 Results

Concentrations for each compounds detected placenta samples with this experiment are illustrated in Figure 3.13 and summarised in Appendix 6. The observed range of concentration was extended from lower than the LOD to 77 ng/g fat at maximum, with almost all values around 0.1 ng g-1 fat. Most of these concentration values appeared 500 times lower than the detectable concentration with non-targeted approach. Thus, those compounds could not be detected with the current NTS method because the concentration is very low in this set of samples. These results were not surprising because the targeted sample preparation was meticulously optimised with different steps of purification to remove all matrix interferences and to concentrate a large volume of sample (12 g) in a small extract of some microliters. Moreover, the applied highly selective MS/MS targeted acquisition mode is more sensitive than the full-scan acquisition used in NTS. However, the concentration detectable by NTS remains in the range of possible concentration levels observed in human samples for highly exposed individuals. As targeted methods are known for their better sensitivity compared to non-targeted methods, this result at an early stage of development is promising to reach sufficient sensitivity with NTS in the coming years to characterise the chemical exposure in human samples.

In order to enhance the method sensitivity and as the instrument LOD is a fixed value, a possible level for action is to introduce an additional concentration factor during the sample preparation, such as to increase the starting sample amount. For the present experiment, only 100 mg were analysed with the non-targeted method whereas 12 g were used for the targeted approach. The effect of the starting sample volume on the physical aspect of the final extract and on analyses was assessed for the Bligh and Dyer method on human milk in the next section.

- 144/296 - Chapter 3. Development of non-targeted sample preparation protocols to screen halogenated markers of chemical exposure in human samples

Figure 3.13: Concentration (ng g -1 lipid) of organic contaminants detected in placenta samples prepared with the targeted method developed by LABERCA and analysed in GC-EI-magnetic sector, LC-ESI-QqQ, LC-ESI-Orbitrap, APGC-QqQ.

- 145/296 - Chapter 3. Development of non-targeted sample preparation protocols to screen halogenated markers of chemical exposure in human samples

Sample amount for analysis

The starting sample volume is generally defined in accordance with accessible quantity and the expected LOD of the method. The high degree of purification of targeted methods allows to purify a large amount of sample and to concentrate compounds of interest. For non-targeted methods where the purification is much more reduced, the starting amount is as low as possible in order to reduce the concentration of matrix interferences. Moreover, for human samples, low volumes are generally available and it entails the use of a low sample amount. However, increasing the starting sample amount can enhance the method sensitivity if matrix interferences do not impair the final detectability of the signals of interest. For the Bligh and Dyer method, three volumes of human milk were treated with proportional volumes of solvent. This experiment was done on human milk instead of placenta as this matrix is accessible in larger volume and was more easily accessible in the context of our work.

In this experiment, 50 µL, 500 µL and 5 mL of human milk, in triplicate, were spiked with 13C- TBBPA and 2H-γ-HBCDD (100 µL at 0.1 ng µL-1), diluted with water (150 µL, 1.5 mL, 15 mL, respectively), extracted by methanol (200 µL, 2 mL, 20 mL, respectively) and chloroform (400 µL, 4 mL, 40 mL, respectively). Samples were vortexed 30 sec after each addition of solvent and finally centrifuged (300 rpm, 10 min, 4 °C). Chloroform fractions were transferred in a vial and dried under a gentle stream of nitrogen at 35 °C. As the applied sample preparation does not remove lipids, a residue of fat was observed for 500 µL and mainly for 5 mL sample amounts. Human milk contains approximately 3 % of fat and this amount was visually observed in the final extract. Extracts were re-suspended in 100 µL of acetonitrile/water 8:2 (v/v) with 2H-α-HBCDD at 0.1 ng µL-1, vortexed and centrifuged and only the supernatant was isolated and injected. For the 5 mL samples, 100 µL were added on the residue of fat in two steps, vortexed and centrifuged and the supernatant was injected.

As illustrated in Figure 3.14, the density of information was higher with 5 mL of milk than with 50 µL. However, the signal intensity of 13C-TBBPA decreased when the milk sample size increased. This decreasing of intensity could be due to matrix effect increasing or/and a fat trapping of CEC with higher sample size inducing a loss of recovery. Anyway, the fat residue observed in the final extract is obviously not compatible with neither LC nor GC analysis.

To conclude, the Bligh and Dyer protocol was found efficient on low volume of matrix (10-100 mg or µL), In order to increase the sensitivity of the method, bigger sample amount can be used,

- 146/296 - Chapter 3. Development of non-targeted sample preparation protocols to screen halogenated markers of chemical exposure in human samples as human milk and placenta are generally available in bigger volume/mass up to 1 mL or g. Thus additional purification step would be required to ensure sufficient compatibility of the final extract with LC/GC analysis. Following this observation, an approach based on the selectivity of a cartridge stationary phase with efficient lipids removal was tested and is presented in the next section 3.4 Captiva EMR-lipid®: application to human milk and meconium.

Figure 3.14: Influence of the sample amount considered for analysis on compounds detection. Overlap of LC-ESI(-)-HRMS total ion currents (top) obtained for a procedural blank (black) and milk samples of 5 mL (green), 500 µL (blue) and 50 µL (red). Overlap of extracted ion chromatograms (bottom) of 13C-TBBPA with the same colour except black, which symbolises the standard in pure solvent. The picture represents the residue of fat observed in the final extract with the three volumes of milk.

3.3.3. Discussion and conclusion

Human placenta samples were analysed to characterise the foetal exposure to organic contaminants with targeted and non-targeted analyses. For the present Ph.D. project, the Bligh

- 147/296 - Chapter 3. Development of non-targeted sample preparation protocols to screen halogenated markers of chemical exposure in human samples and Dyer protocol was assessed to prepare human samples according to a non-targeted approach. Samples were refined with a single liquid-liquid extraction where almost all compounds were extracted and partitioned between a polar and nonpolar fraction. This method offers satisfactory results with recovery between 20 and 70 %, after the addition of signals detected in both polar and nonpolar fractions. In addition, most of standards were detected in the nonpolar fraction which demonstrated the need to investigate this fraction first.

The bottleneck of non-targeted screening methods is not only the data processing but also the sensitivity of the method. In order to assess the compatibility of the protocol to detect compounds in low concentrations, samples were also analysed with an optimised targeted method as well. It is already assumed that NTS methods will not reach the same sensitivity as targeted methods, but it is important to verify the adequacy of the method with the final purpose. In the present case, the method can detect compounds in a plausible range of concentration for highly exposed individuals. As the LOD is partly defined by the instrument, an option to reach lower LOD would be to use a more sensitive instrumentation. For the present study, the last generation of detectors was used (Orbitrap), powerful enough to reach low LOD and provide high mass resolution. At the present stage, users cannot directly improve the sensitivity of the instrument and should work on the sample preparation. For instance, increasing the initial volume enables to concentrate analytes of interest. However, the sample preparation should be sufficiently elaborate to remove matrix interferences while preserving organic contaminants. This option was assessed and the Bligh and Dyer does not remove matrix interferences, thus the initial sample volume should be low. In that context, another approach was also tested to remove lipids and is presented in the next section.

3.4. Captiva EMR-lipid®: application to human milk and meconium

3.4.1. Introduction

The sample preparation of biological matrices with high lipid content is known as challenging because a wide range of contaminants have lipophilic properties (e.g. POPs) which make the selective extraction of compounds of interest without lipids particularly difficult. Except for lipidomics studies, lipids are considered as matrix interferences and should be removed to avoid the degradation of the overall performance of the method, which also makes data processing

- 148/296 - Chapter 3. Development of non-targeted sample preparation protocols to screen halogenated markers of chemical exposure in human samples more complex. Targeted methods are normally developed with several purification steps to remove those lipids. Moreover, most of methods are based on the retention of targeted compounds on a stationary phase, such as SPE. These two aspects are not in adequacy with the NTS, as a large range of compounds is sought with minimal sample preparation and the range of expected compounds is too large to be retained on a single stationary phase. However, an opposite approach, based on the selective retention of matrix interferences, is compatible with NTS. Cartridges with different stationary phases provide efficient lipids removal have recently been proposed by different private companies. Agilent sells a polymeric stationary phase with efficient lipids removal by hydrophobic interaction and size exclusion. However, the exact mode of action (phase under patent) remains imperfectly described. In theory, it should purify the sample while preserving exogenous contaminants of interest (Han et al., 2016; Zhao et al., 2019; Arce-Lopez et al., 2020).

In the context of the present Ph.D. work, the Captiva EMR-lipid® cartridge, with lipid-specific retention, was tested on human milk and meconium in order to remove lipids while preserving organic contaminants. The comprehensive characterisation of samples was supported by a dual detection LC- and GC-HRMS to maximise the range of accessible markers of exposure.

During the present Ph.D. project, a workflow based on the same guidelines as for placentas samples was developed to characterise real human milk and meconium samples with a non-targeted approach. A method performance assessment procedure using advanced and predictive modelling tools was also established.

3.4.2. Method development

Protocol adaptation to human milk

 Protocols

Human milk is a complex matrix composed of water, lipids, proteins and other nutrients (Petherick, 2010), thus requiring dedicated sample preparation. Three protocols were developed to assess and compare the compatibility of Captiva EMR-lipid® cartridge. The cartridge was used alone (method A), in combination with LLE (method B), in combination with LLE and SPE (method C)(Figure 3.15). The polymeric stationary phase Captiva EMR-lipid® is stable when samples contain 20 % water by volume in combination with organic solvent (acetonitrile, methanol). On the contrary, the Captiva EMR-lipid® phase is desegregated with other organic

- 149/296 - Chapter 3. Development of non-targeted sample preparation protocols to screen halogenated markers of chemical exposure in human samples solvents as dichloromethane. The portion of water was considered in milk because it naturally contains ~90 % water by volume (Petherick, 2010). Therefore, in the further described experiments, the 20 % of water were actually milk, and the 80 % of organic solvent were acetonitrile.

Figure 3.15: Three protocols tested to prepare human milk based on proteins and lipids removal.

All glassware was burned at 400 °C overnight to minimise external contamination. A pool of human milk, stored at -20 °C, was constituted from five human milk samples (4 mL each) collected in the frame of a previous research project and used for all our development experiments. It was thawed and vigorously shaken before pipetting. For the comparison of methods A, B and C, 100 µL of milk were spiked with 50 µL of 13C-TBBPA and 2H-γ-HBCDD (0.1 ng µL-1) and 50 µL of Cl/Br-phenolic compounds of standard mix 1 (0.1 ng µL-1). Proteins were denatured with 400 µL of acetonitrile. Samples were vortexed for 30 sec and centrifuged for 10 min at 4 °C and 3000 rpm. The supernatant was loaded on the Captiva EMR-Lipid® (1 mL, 40 mg) stationary phase, previously conditioned with 2*800 µL of acetonitrile/water 8:2 (v/v). After the elution at atmospheric pressure without additional solvent, eluates were either i) evaporated under a gentle stream of nitrogen and reconstituted in 50 µL of acetonitrile/water 8:2 (v/v) containing 2H-α-HBCDD (0.1 ng µL-1) (method A); ii) extracted twice with 500 µL of hexane, from which fractions were evaporated under a gentle stream of nitrogen and reconstituted in 50 µL of acetonitrile/water 8:2 (v/v) for the polar fraction and in 50 µL of

- 150/296 - Chapter 3. Development of non-targeted sample preparation protocols to screen halogenated markers of chemical exposure in human samples acetonitrile for the nonpolar fraction, both containing 2H-α-HBCDD (0.1 ng µL-1) (method B); iii) extracted twice with 500 µL of hexane, then the polar fraction was purified with a SPE C18 (1g, 6 mL) (conditioning: 5 mL of methanol, 5 mL of water, rinsing 5 mL water, elution: 2*5 mL methanol) and the nonpolar fraction was purified with a SPE SiOH (1g, 6 mL) (conditioning: 5 mL of dichloromethane, 5 mL of cyclohexane, rinsing 5 mL cyclohexane, elution: 2*5 mL dichloromethane). Following SPE, each fraction was evaporated under a gentle stream of nitrogen and reconstituted, in 50 µL of acetonitrile/water 8:2 (v/v) for the polar fraction and in 50 µL of acetonitrile for the nonpolar fraction, both containing 2H-α-HBCDD (0.1 ng µL-1) (method C).

All extract from methods A, B and C were then analysed in LC-HRMS with optimised parameters detailed in Chapter 2 (LC gradient G1).

 Method comparison criteria

Repeatability was assessed as the relative standard deviation (RSD) of the signal intensities observed for each Cl/Br-phenolic compound of the standard mix 1, in triplicate samples. Phenol responses were summed by sub-family (e.g. Chlorophenol signal was the sum of signals of 2- chlorophenol + 3-chlorophenol + 4-chlorophenol).

For each compound, recovery was calculated as detailed bellow:

퐴푟푒푎 + 퐴푟푒푎 푅푒푐표푣푒푟푦 (%) = ( 푃퐹 푁푃퐹) ∗ 100 퐴푟푒푎푆푡푎푛푑푎푟푑

Where AreaPF is the signal response observed for the spiked compound in the polar fraction,

AreaNPF is the signal response observed for the spiked compound in the nonpolar fraction, and

Areastandard is the signal response observed for the analytical standard in pure solvent.

 Results

Firstly, polar fractions resulting from methods A, B and C were compared on a macroscopic view. As illustrated in Figure 3.16, the addition of purification step decreases the intensity of the total ion current. This implies that the concentration of compounds and/or matrix interferences also decreases. This could probably reduce the risk of ion suppression.

- 151/296 - Chapter 3. Development of non-targeted sample preparation protocols to screen halogenated markers of chemical exposure in human samples

Figure 3.16: Total ion currents obtained for human milk treated by sample preparation method A (red), B (blue) and C (green), and a procedural blank (black) analysed in LC-ESI(-)-HRMS.

As regard to a more detailed analysis, RSD (Table 3.2) of the signal intensity detected for each compound, for polar fraction resulting from methods A, B and C were found globally similar and lower than 30%. Conversely, RSD observed for nonpolar fractions resulting from method C (> 50 %) were found much higher than those of method B (30 %), probably due to lower concentrations detected with method C. Mono-chloro/bromo-phenols and HBCDD were not detected with any of the applied method. Those compounds then appeared as out of the application range of the applied methods and complementary approaches should be developed to analyse them. For HBCDD, the signal was probably suppressed by matrix effect as with a linear gradient of elution (G3) RSD and recovery of this compound were acceptable (see section 3.4.3 Method assessment).

- 152/296 - Chapter 3. Development of non-targeted sample preparation protocols to screen halogenated markers of chemical exposure in human samples

Table 3.2: Repeatability (RSD in %) observed for Cl/Br-phenolic compounds spiked in 100 µL of human milk prepared with method A, B and C and analysed by LC-ESI(-)-HRMS. frac: Fraction; ND: not detected.

Method B Method B Method C Method C Method A polar frac. nonpolar frac. polar frac. nonpolar frac. Chlorophenols ND 51% ND ND ND Dichlorophenols 9% 16% 23% 27% 53% Trichlorophenols 19% 13% 16% 24% 45% Tetrachlorophenols 22% 18% 14% 24% 52% Pentachlorophenol 16% 14% 9% 26% 45% Bromophenols 18% 36% ND ND ND Dibromophenols 14% 13% 15% 20% 50% Tribromophenols 23% 12% 13% 20% 53% Tetrabromophenols 25% 12% 4% 27% 45% Pentabromophenol 27% 15% 16% 26% 21% HdCteBDE 6% 22% 8% 30% 46% HhBDE 18% 14% 2% 43% 50% HtCDE 36% 14% 9% 25% 4% HtBDE 25% 15% 3% 23% 5% HCteBDE 16% 8% 4% 32% 52% HpBDE 11% 12% 11% 19% 50% 13C-TBBPA 23% 10% ND 17% ND 2H-g-HBCDD ND ND ND ND 24 %

Finally, recoveries observed with method B were between 40 and 60%, thus higher than recoveries obtained with methods A and C, both between 20 and 40% (Figure 3.17). This can be explained by higher matrix effect with the method A, which decreases with additional purification steps in method B, and low recovery with method C because of a too selective preparation. Recoveries observed for mono- and di- chloro/bromo-phenol should not be considered as they were out of the range (> 100%). This confirms that those compounds are part of the method limitation.

To conclude, method B was a compromise to eliminate matrix interferences while preserving the detectability of most of the Cl/Br-phenolic compounds of the standard mix 1.

- 153/296 - Chapter 3. Development of non-targeted sample preparation protocols to screen halogenated markers of chemical exposure in human samples

Figure 3.17: Mean recovery (n=3) observed for some of Cl/Br-phenolic compounds of the standard mix 1 after application of the sample preparation method A, B and C and detection by LC-ESI(-)-HRMS.

- 154/296 - Chapter 3. Development of non-targeted sample preparation protocols to screen halogenated markers of chemical exposure in human samples

Sample volume for analysis

 Protocol

Starting with higher volume of sample may enable to reach lower detection limits due to the possible concentration factor through the applied sample preparation procedures. However, matrix interferences should be removed to avoid any ion suppression occurring during the ionisation. As it was demonstrated with the Bligh and Dyer approach, the starting volume of samples can influence the compatibility of the final extract with LC- and GC- injection. According to the previous development, the optimised method (method B) can treat 100 µL of human milk. It was assumed that proportional volumes of solvent to denature proteins and size of the stationary phase to remove lipids allowed to increase the starting volume of human milk. Two higher volumes of milk (400 µL and 500 µL) were tested with proportional volume of acetonitrile (1.6 mL and 2 mL, respectively) and size of Captiva EMR-Lipids® cartridge (6 mL, 600 mg for both, whereas a cartridge featuring 1 mL, 40 mg was handle to prepare 100 µL of milk). Samples in triplicate were spiked with the QA/QC mix 2 (25 ng/sample). Eluates were extracted with 2*2 mL and 2*2.5 mL of hexane, respectively. The rest of the protocol was identical for both volumes of samples. Samples were analysed in LC-ESI(+/-)-HRMS (elution gradient G3) and GC-EI-HRMS according to the method developed in Chapter 2 (sections 2.2.2 and 2.2.3).

 Method comparison criteria

It was assumed that the bigger volume of acetonitrile used to precipitate proteins contributed to elute lipids normally caught on the stationary phase. It was also expected that recovery and matrix effect could be higher with a larger volume of acetonitrile as more compounds of interest and matrix interference could be eluted. The first criterion assessed was the physical aspect of final extracts and their compatibility with LC and GC injections. According to this hypothesis, recovery and matrix effect were calculated with equation 1 and 2, respectively. Matrix effect is commonly judged acceptable in targeted analysis if the signal variation is lower than 30 %.

Equation 1:

퐴푟푒푎 − 퐴푟푒푎 푅푒푐표푣푒푟푦 (%) = ( 퐵퐸 푈푆) ∗ 100 퐴푟푒푎푆푡푎푛푑푎푟푑

- 155/296 - Chapter 3. Development of non-targeted sample preparation protocols to screen halogenated markers of chemical exposure in human samples

Equation 2:

퐴푟푒푎 − 퐴푟푒푎 푀푎푡푟푖푥 푒푓푓푒푐푡 (%) = ( 퐴퐸 푈푆 − 1) ∗ 100 퐴푟푒푎푆푡푎푛푑푎푟푑

Where AreaBE and AreaAE are the signal response observed for the spiked compound before and after the extraction respectively, AreaUS is the signal response observed for the compound in extracted and unspiked sample (if it was detected) and Areastandard is the signal response observed for the analytical standard in pure solvent.

 Results

The first difference observed between the experiments with the various sample amounts considered for analysis consisted in the physical aspect of the final extract of the polar phase. 500 µL of milk led to a residue of fat in the polar fraction vial which was not observed with 400 µL of milk. During the LC-HRMS analysis, no carry-over was observed, and it could be attributed to the included maintenance step, where the column is flushed after each injection. However, the presence of a residue in the extract is not compatible with the chromatographic system and could induce matrix effect. Thus, 400 µL appeared more appropriate for the present study.

Fat residue was not observed in the nonpolar fraction. However, important carry-over during the GC-HRMS analysis with high density in TIC were detected. In Figure 3.18, TIC for both samples of 400 µL and 500 µL of milk are illustrated as well as the TIC observed for a pure solvent sample injected just after the previous milk sample extracts. On one hand, the TIC observed for a 500 µL milk sample globally appear more informative than the one obtained for a 400 µL. On the other hand, high carry-over was observed on the injection of solvent right after the sample of 500 µL, which was not the case for the 400 µL of milk. As for LC analysis, 500 µL of milk sample volume were then estimated to be not compatible with the GC system.

- 156/296 - Chapter 3. Development of non-targeted sample preparation protocols to screen halogenated markers of chemical exposure in human samples

Figure 3.18: Total ion current obtained for 400 µL and 500 µL human milk treated with the Captiva EMR-Lipid® protocol and corresponding procedural blanks analysed by GC-EI- HRMS.

Recovery and matrix effect were assessed for both tested sample amounts (Figure 3.19) only for standards (QA/QC mix 2) detected in LC-HRMS since results in GC-HRMS could not be accurately investigated because of the carry-over. Obtained recoveries were higher with 500 µL than 400 µL of sample for some compounds (e.g. TCPy, fenvalerate free acid) whereas it was the contrary for some others (e.g. TBBPA, a-HBCDD). Matrix effect was in the range ± 30% for 400 µL of sample whereas it was partly out of it for 500 µL of milk. Finally, 400 µL of milk appeared as a good compromise to limit the matrix effect while maintaining sufficient recoveries.

To conclude, it is possible to prepare up to 400 µL of milk with the present protocol. A detailed protocol is available in Appendix 7.

- 157/296 - Chapter 3. Development of non-targeted sample preparation protocols to screen halogenated markers of chemical exposure in human samples

Figure 3.19: Mean recovery (top) and matrix effect (bottom) (n=3) observed for QA/QC mix 2 compounds spiked in 400 µL (red) and 500 µL (blue) of human milk treated with the Captiva EMR-Lipid® approach analysed in LC-ESI(-)-HRMS.

Protocol transposition to meconium samples

In collaboration with the Research Centre for Toxic Compounds in the Environment (RECETOX, Masaryk University, Brno, Czech Republic), the protocol developed to prepare human milk was transposed to meconium during a secondment in RECETOX.

Naturally, milk is a liquid matrix with hydrophilic and lipophilic properties, thus extraction by an organic solvent is a known approach compatible with the matrix. Meconium is a solid and

- 158/296 - Chapter 3. Development of non-targeted sample preparation protocols to screen halogenated markers of chemical exposure in human samples viscous matrix composed of water, lipids, proteins, bile acids, amniotic fluid residues and other biological substances (enzymes, bacteria) (Cassoulet et al., 2019). The addition of acetonitrile in a single step could lead to the aggregation of the sample and reduce extraction efficiency. In order to determine the most suitable approach, three extractions were performed in triplicate on 500 mg of the fresh meconium sample as follows:

- Extraction 1: sample was first dissolved in 400 µL of water and then 1.6 mL of acetonitrile were added

- Extraction 2: sample was first dissolved in 800 µL of water and then 3.2 mL of acetonitrile were added

- Extraction 3: sample was directly dissolved in 1.6 mL of acetonitrile

Extracts were purified following the same protocol as developed for human milk. Samples were analysed in GC-HRMS at RECETOX with the same conditions to those optimised in Chapter 2. The single difference is the connection of a guard column (Rxi® Guard Column, 2 m x 0.53 mm) to the analytical column (Rxi-5Sil MS, 30 m x 0.25 mm, 0.25 µm) for GC injection performed at RECETOX.

Meconium was found well dissolved in water with method 1 and 2, and it was aggregated after the addition of acetonitrile, whereas it directly precipitated in acetonitrile with extraction 3 and was stuck on glass (Figure 3.20). The extraction 3 was physically more difficult because the contact between sample and solvent was reduced. After the sample purification on Captiva EMR-Lipid® cartridge and partitioning with hexane, extracts obtained with the second method were more coloured than others. Indeed, following evaporation of both polar and nonpolar fractions, a residue was observed at the bottom of the vial in the polar fraction. The second extraction used a larger volume of solvent than extraction 1 and 3 and it probably eluted lipids caught on the stationary phase resulting in a residue of fat after evaporation (as previously demonstrated with human milk samples).

- 159/296 - Chapter 3. Development of non-targeted sample preparation protocols to screen halogenated markers of chemical exposure in human samples

Figure 3.20: Meconium (500 mg) sample preparation with 3 extractions approaches E1: 400 µL water + 1.6 mL acetonitrile; E2: 800 µL water + 3.2 mL acetonitrile; E3: 1.6 mL acetonitrile and purified with the Captiva EMR-Lipid® cartridge.

All hexane fractions and procedural blanks were analysed by GC-HRMS. On total ion currents (TIC), more peaks with higher intensities were detected with extraction 2, then extraction 1 and finally extraction 3, in a descending order. This is in agreement with the larger volume of solvent used to extract more compounds and also because the purification step did appear insufficient for those samples. According to these observations, the second extraction was not selected and extraction 1 was retained for further investigation because it is a compromise for homogenising the sample through facilitated extraction with a first dissolution in water and handling an appropriate volume of solvents compatible with Captiva EMR-Lipid® stationary phase retention.

- 160/296 - Chapter 3. Development of non-targeted sample preparation protocols to screen halogenated markers of chemical exposure in human samples

Sample amount for analysis

As for human milk, three starting sample amounts (200, 400 and 500 mg) were tested in triplicate. Meconium samples were first dissolved in water (400 µL) and proteins were precipitated in acetonitrile (1.6 mL) containing standards of the QA/QC mix 2 (5 ng/sample). Extracts were analysed in LC-ESI(+/-)-HRMS and GC-EI-HRMS with the same optimised conditions as developed in Chapter 2. It was assumed that spiking a solid matrix is less representative of the extraction efficiency than for a liquid matrix. For the present development, the implementation of standards in the experiment was used in order to estimate the compatibility of the sample amount with the detection mode. Signals of standards were used to approximate the matrix effect according to the sample amount rather than recoveries.

The total ion current obtained for the three sample amounts appeared very similar, so all 200, 400 and 500 mg sample amounts were estimated to be compatible with the sample preparation protocol and the chromatographic system (Figure 3.21). Then, the signal of each standard was assessed according to the sample amount (Figure 3.22) and the intensity was very similar for the three sample volumes. Thus, increasing the amount of sample does not increase the matrix effect. Based on these results, 500 mg were finally selected to analyse real samples of meconium.

- 161/296 - Chapter 3. Development of non-targeted sample preparation protocols to screen halogenated markers of chemical exposure in human samples

Figure 3.21: Total ion currents obtained for 200 mg (red), 400 mg (blue) and 500 mg (black) of meconium prepared with the Captiva EMR-Lipid® protocol and analysed by LC-ESI(-)-HRMS.

- 162/296 - Chapter 3. Development of non-targeted sample preparation protocols to screen halogenated markers of chemical exposure in human samples

Figure 3.22: Repeatability of extractions for 200 mg (grey), 400 mg (orange) and 500 mg (red) of meconium treated with the Captiva EMR-Lipid® protocol and analysed by LC-ESI(+/-)-HMRS and GC-EI-HRMS.

Additional rinsing step

The Captiva EMR-Lipid® phase was used in the cartridge design and not in dispersive powder mode. Elution was done under atmospheric pressure. It is an intrinsic and physical property that the solvent (i.e. potentially compounds of interest) is kept on the stationary phase after the elution if no additional positive or negative pressure is applied. Also, an additional step of rinsing with solvent could release those compounds dissolved in the solvent. As previously

- 163/296 - Chapter 3. Development of non-targeted sample preparation protocols to screen halogenated markers of chemical exposure in human samples illustrated for milk and meconium analyses, too large a volume of solvent loaded on the Captiva EMR-Lipid® was suspected not only to elute compounds of interest but also matrix interferences. An experiment was conducted to study this hypothesis. For the same samples as above (triplicate of 200, 400 and 500 mg of meconium), the cartridge was rinsed after the main elution with 200 µL of acetonitrile/water 8:2 (v/v) at atmospheric pressure and collected in a separated vial. The solvent was evaporated, then extracts were reconstituted in acetonitrile/water 8:2 (v/v) and analysed in LC-ESI(-)-HRMS. The TIC obtained for the second elution was less intense that the TIC of the first elution but still very intense. In the present case, extracts were separately analysed, but the potential purpose would be the combination of the second elution with the first one. Consequently, the TIC could be denser and the extract may not be compatible with the system of injection. In order to estimate the potential gain offered by a second elution, recoveries for each standard for the two successive operated extractions were calculated. The obtained results are illustrated in Figure 3.23 with a gain from the second elution, illustrated as a positive error bar, lower than 5%. To conclude, the additional recovery offered by a second elution step appeared lower than 50 % of the initial value. Also, adding a second step of elution could lead to incompatibility of the extract with LC and GC analyse due to insufficient purification. This additional rinsing step was then not retained for further analysis. According to this last optimisation, a detailed protocol is available in Appendix 7.

Figure 3.23: Mean recoveries observed (n=3) for 200 mg (grey), 400 mg (orange) and 500 mg (red) of meconium treated with the same protocol and analysed by LC-ESI(+/-)-HMRS and GC-EI-HRMS. Diagram bars represent the recovery after the first elution and positive error bars the gain with the second elution.

- 164/296 - Chapter 3. Development of non-targeted sample preparation protocols to screen halogenated markers of chemical exposure in human samples

An alternative to reduce the loss of compounds could be the dispersive powder design of Captiva EMR-Lipid®. This would allow determining whether matrix interferences were eluted thanks to the large volume of solvent or because the stationary phase was not able to catch those compounds through chemical interactions.

Liquid-liquid partitioning optimisation

Following the purification on Captiva EMR-Lipid®, the developed sample preparation includes a liquid-liquid partitioning, which was a matter of a separate assessment. This partitioning was conducted on 6 replicates where 50 µL of QA/QC mix 2 were diluted into 2 mL of acetonitrile/water 8:2 (v/v) and extracted twice with 2 mL of hexane. Each fraction was evaporated until dryness under a gentle stream of nitrogen. Three polar fractions and the corresponding three nonpolar fractions were reconstituted in acetonitrile/water 8:2 (v/v) and analysed in LC-HRMS. The six last fractions (three polar and three nonpolar) were reconstituted in hexane and injected in GC-HRMS. The result of the partitioning is illustrated in Figure 3.24. The obtained results globally confirmed that compounds preferably detected in LC are in the polar fraction whereas compounds detectable in GC are in the nonpolar fraction. Some exceptions are highlighted with acetochlor and metolachlor, both partitioning in the two fractions and also both detectable in LC and GC with a similar limit of detection. For those compounds, the double detection should be considered for semi-quantitative approach (e.g. to estimate method recovery).

In the scope of a comprehensive characterisation of a sample, the ideal situation would be to analyse each fraction with their respective optimal detection mode, leading to a total of six injections per sample, approximately 4.5 hours of analysis and a laborious data processing. This is feasible for few samples but not realistic in a context of high throughput human biomonitoring research to support large-scale biomonitoring studies. According to the results of the present study on the liquid-liquid partitioning, the analysis of the polar fraction in LC- HRMS and the nonpolar fraction in GC-HRMS already enables to detect a large range of markers. This complementarity of techniques combined with an appropriate sample preparation method then appears as a good strategical option to characterise samples with a non-targeted approach.

- 165/296 - Chapter 3. Development of non-targeted sample preparation protocols to screen halogenated markers of chemical exposure in human samples

Figure 3.24: Liquid-liquid partitioning of the QA/QC mix 2 compounds between polar (acetonitrile/eau in blue) and nonpolar (hexane in red) phases detected in LC-ESI(+/-)-HRMS (top and middle left) and GC-EI-HRMS (middle right and bottom). *: compounds detected in GC but preferably analysed in LC.

- 166/296 - Chapter 3. Development of non-targeted sample preparation protocols to screen halogenated markers of chemical exposure in human samples

3.4.3. Method assessment

The method performance was assessed in order to determine its strengths and weaknesses regarding compounds of the QA/QC mix 2.

A pool of human milk, composed of 2.5 mL of five different milk, was used for this experiment and 400 µL of milk were prepared in triplicate for each condition:

- Unspiked samples - Spiked samples before the sample preparation with standards of the QA/QC mix 2 and atrazine and 13C-PFOS at 6.25; 12.5; 62.5; 125 pg µL-1 milk - Spiked samples after the sample preparation with standards of the QA/QC mix 2 at 62.5 pg µL-1 milk

Three procedural blanks where matrix was replaced by 400 µL of water were also prepared. All samples, including procedural blanks were internally spiked with 13C-HBBz, 2H-α-HBCDD, 2H-BDCIPP and externally spiked with 13C-anti-DP for the nonpolar fraction and 2H-γ-HBCDD, 13C-TBBPA and 13C-atrazine for polar fraction (5 ng each).

Method efficiency was assessed based on five classical criteria including sensitivity (previously described in Chapter 2), linearity, repeatability, recovery and matrix effect (ME), using the standards of the QA/QC mix 2. A thorough analysis of the observed method performance, based on a multivariate regression model (Orthogonal Partial Least Squares OPLS), was conducted to relate the signal intensities obtained for a range of reference standard compounds to their individual physicochemical properties.

Linearity

The linear calibration curve of four concentration levels (6.25; 12.5; 62.5; 125 pg µL-1 human milk) was accepted if the coefficient of determination (R2) was greater than 0.99.

Linearity was estimated as the correlation between the peak area and the concentration in spiked samples, was found satisfying, with R2 values higher than 0.99 for all analysed compounds either in GC- or LC-HRMS, except for α-HBCDD, deltametrin and BTBPE for which R2 was better than 0.97 (Table 3.3).

- 167/296 - Chapter 3. Development of non-targeted sample preparation protocols to screen halogenated markers of chemical exposure in human samples

Repeatability

Repeatability was assessed as the relative standard deviation (RSD) of the signal intensities observed for each compound and each concentration level in triplicate. The repeatability was considered acceptable if RSD < 30%.

Repeatability was globally concluded to be satisfying with RSD lower than the fixed limit of 30% for most of the compounds (Table 3.4).

Recovery

Recovery was calculated with equation (1) for all concentration levels with the instrumental method offering the lowest LOD (see Chapter 2). According to the study of the liquid-liquid partitioning, acetochlor and metolachlor are split into both fractions and detected in LC and GC. For those two compounds, the recovery was calculated as the sum of both recoveries calculated for LC and GC analysis.

Equation 1: 퐴푟푒푎 − 퐴푟푒푎 푅푒푐표푣푒푟푦 (%) = ( 퐵퐸 푈푆) ∗ 100 퐴푟푒푎푆푡푎푛푑푎푟푑

With AreaBE the response observed for the spiked compound before the extraction, AreaUS the response observed for the compound in unspiked sample (if it was detected) and Areastandard is the signal response observed for the analytical standard in pure solvent.

As shown in Table 3.3, most of compounds (86%, 100% and 57%) were well recovered respectively through the LC-ESI(-), LC-ESI(+) and GC-EI methods. Molecules with recovery lower than 10% allow to draw methods limitations, as for 2,4-DCP, 2,4-DBP, p-TBX, PBDE 153, BTBPE, and anti-DP.

Matrix effect

Matrix effect was calculated according to equation (2) at the concentration level 62.5 pg µL-1 human milk, in triplicate. The result has been evaluated on scientist opinion, taking into account its consequence with regard to the repeatability and sensitivity.

- 168/296 - Chapter 3. Development of non-targeted sample preparation protocols to screen halogenated markers of chemical exposure in human samples

Equation 2: 퐴푟푒푎 − 퐴푟푒푎 푀푎푡푟푖푥 푒푓푓푒푐푡 (%) = ( 퐴퐸 푈푆 − 1) ∗ 100 퐴푟푒푎푆푡푎푛푑푎푟푑

With AreaAE the response observed for the spiked compound after the extraction respectively and AreaUS the response observed for the compound in unspiked sample (if it was detected) and

Areastandard is the signal response observed for the analytical standard in pure solvent.

As shown in Table 3.3, no significant matrix effect was observed with regard to the LC-ESI(+/-)-HRMS detection, since no value exceeded 30%, and the purification strategy was concluded satisfactory. Conversely, the GC-EI-HRMS detection was influenced by significant matrix effect, HCB, p-TBX, HBBz diagnostic signals being decreased between 40 and 60%. On the contrary, quizalofop-p-ethyl and deltametrin signals were increased of 216 and 95%, respectively. Those signal fluctuations could probably be reduced with more purified extracts to decrease matrix effect but it would lead to more selective sample preparation which is contrary to non-targeted approaches. New sample preparation approaches have then to be elaborated for NTS, which take into account this apparent contradictory objective to encompass an extended range of biomarkers of exposure while preserving the compatibility of the resulting extracts with the measurement systems. This imposes in particular to implement new preventive maintenance dispositions as guard-column and/or pre-filter and either in LC or GC. In LC a regular column clean-up after each injection is recommended, in the present case with isopropanol/acetone 1/1 (v/v), to ensure system robustness and prevent peak resolution degradation.

- 169/296 - Chapter 3. Development of non-targeted sample preparation protocols to screen halogenated markers of chemical exposure in human samples

Table 3.3: Sample preparation linearity and matrix effect (ME) in GC-EI-HRMS and LC-ESI(+/-)-HRMS, and total recovery at four concentration levels (level 1 to 4 = 6.25; 12.5; 62.5; 125 pg µL-1). ND: not detected.

GC-EI LC-ESI(-) LC-ESI(+) Recovery Compounds name R2 ME R2 ME R2 ME Level 1 Level 2 Level 3 Level 4

2,4-DCP - - ND -16% - - ND ND ND ND TCPy - - 0.992 +11% - - 34% 27% 35% 32% Simazine - - - - 0.999 -5% 33% 28% 31% 31% Fenvalerate free acid - - 0.992 +6% - - 11% 9% 24% 21% 2,3,4,5-tetra-CP - - 0.993 +6% - - 20% 13% 24% 24% 2,4-DBP - - 0.997 +23% - - ND ND 2% 3% Acetochlor 0.997 -23% - - 0.997 0% 28% 22% 39% 33% HCB 0.994 -58% - - - - 6% 2% 5% 7% Metolachlor 0.997 -25% - - 0.995 -2% 33% 24% 39% 32% β-HCH 0.998 -23% - - - - 5% 3% 3% 6% Triclosan - - 0.994 +26% - - 40% 28% 40% 37% Fenhexamid - - 0.996 +3% - - 40% 36% 44% 42% p,p'-DDE 0.989 -12% - - - - 23% 11% 17% 24% Chlorpyrifos 0.998 -10% - - - - 39% 20% 20% 38% Chlorfenvinphos - - - - 0.999 -2% 29% 21% 28% 28% Tetraconazole - - 0.995 +2% 1.000 +3% 37% 27% 38% 35% Quizalofop-p-ethyl 0.998 +216% - - - - 31% 26% 10% 16% Prochloraz - - - - 0.999 0% 17% 12% 16% 17% (Z)-Dimetomorph - - - - 0.999 +2% 33% 24% 32% 33% 2,3,4,5-tetra-BP - - 0.996 0% - - 27% 23% 31% 32% p-TBX 0.990 -51% - - - - 2% 0% 1% 1% Fipronil - - 0.997 -3% - - 36% 29% 38% 37% Deltametrin 0.972 +95% - - - - 20% 22% 5% 12% TBBPA - - 0.997 +25% - - 29% 28% 41% 43%

- 170/296 - Chapter 3. Development of non-targeted sample preparation protocols to screen halogenated markers of chemical exposure in human samples

HBBz 0.991 -40% - - - - 2% 2% 4% 5% α-HBCDD - - 0.973 +21% - - 12% 12% 28% 23% PBDE 153 0.996 -29% - - - - 1% 0% 1% 2% anti-DP 0.990 -29% - - - - 0% 0% 0% 0% OH-BDE 137 - - 0.998 -6% - - 20% 19% 26% 28% BTBPE 0.978 +19% - - - - 0% 0% 0% 0%

Table 3.4: Signal variability (relative standard deviation (RSD) observed for all compounds in triplicate in different detection modes, spiked before the extraction at four concentration levels (level 1 to 4 = 6.25; 12.5; 62.5; 125 pg µL-1) and spiked after extraction (spiked after ext.) at 62.5 pg µL-1. *RSD on duplicate sample.

GC-EI (RSD) LC-ESI(-) (RSD) LC-ESI(+) (RSD) Compounds name Spiked Spiked Spiked Level 1 Level 2 Level 3 Level 4 Level 1 Level 2 Level 3 Level 4 Level 1 Level 2 Level 3 Level 4 af. ext. af. ext. af. ext. 2,4-DCP - - - - - ND ND ND ND 24% - - - - - TCPy - - - - - 17% 19% 2% 16% 3% - - - - - Simazine ------18% 16% 3% 18% 1% Fenvalerate free acid - - - - - 87% 103% 7% 23% 3% - - - - - 2,3,4,5-tetra-CP - - - - - 32% 48% 10% 30% 2% - - - - - 2,4-DBP - - - - - ND ND 28% 74% 10% - - - - - Acetochlor 16% 4% 10% 10% 5% - - - - - 25% 45% 5% 27% 2% HCB 22% 50% 23% 26% 35% ------Metolachlor 16% 5% 9% 8% 6% - - - - - 22% 28% 6% 26% 4% b-HCH 13% 2% 10% 0% 8% ------Triclosan - - - - - 34% 26% 12% 20% 5% - - - - Fenhexamid - - - - - 34% 18% 9% 16% 21% - - - - - p,p'-DDE 25% 13% 6% 0% 4% ------Chlorpyrifos 25% 38% 4% 37% 8% ------Chlorfenvinphos ------23% 25% 6% 23% 2%

- 171/296 - Chapter 3. Development of non-targeted sample preparation protocols to screen halogenated markers of chemical exposure in human samples

Tetraconazole - - - - - 32% 18% 12% 21% 1% 25% 22% 9% 25% 4% Quizalofop-p-ethyl 12% 6% 16% 6% 5% ------Prochloraz ------32% 37% 24% 47% 4% (Z)-Dimetomorph ------21% 23% 7% 27% 2% 2,3,4,5-tetra-BP - - - - - 26% 15% 11% 15% 1% - - - - - p-TBX 45% 37% 27% 17% 13% ------Fipronil - - - - - 24% 16% 8% 13% 4% - - - - - Deltametrin 76% 25% 36% 83% 13% ------TBBPA - - - - - 38% 22% 17% 19% 5% - - - - - HBBz 98% 35% 19% 7% 7% ------α-HBCDD - - - - - 63% 41% 2% 15% 16% - - - - - PBDE-153 117% 141% 28% 58% 2% ------anti-DP 91% 141% 46% 44% 2% ------OH-BDE-137 - - - - - 36% 15% 9% 12% 6% - - - - - BTBPE ND ND 65% 63% 6% ------

- 172/296 - Chapter 3. Development of non-targeted sample preparation protocols to screen halogenated markers of chemical exposure in human samples

Finally, all the test reference compounds passed the qualitative detection criterion at different concentration levels, except 2,4-DCP which was not detected and 2,4-DBP and BTBPE which were only detected at the two highest concentration levels. A subset group of 19 compounds also passed the linearity, ME and recovery criteria as summarised in Table 3.5. In order to better understand the root causes of these method limitations in terms of accessible markers, a statistical modelling approach was conducted to predict the expected recovery from those marker’s physicochemical properties.

Table 3.5: Subset group of reference test compounds with validated criteria. LOD: Limit of detection in ng µL-1, ME: matrix effect.

Compounds name LOD Linearity ME

Acetochlor 0.005 0.997 -23%

EI Metolachlor 0.01 0.997 -25%

-

GC Chlorpyrifos 0.005 0.998 -10% p,p'-DDE 0.001 0.989 -12% 2,3,4,5-tetra-CP 0.001 0.993 +6% 2,3,4,5-tetra-BP 0.005 0.996 0% Tetraconazole 0.001 0.995 +2% Fipronil 0.001 0.997 -3%

) - OH-BDE 137 0.001 0.998 -6%

ESI( TBBPA 0.005 0.997 +25%

-

LC α-HBCDD 0.01 0.973 +21% TCPy 0.01 0.992 +11% Fenvalerate free acid 0.1 0.992 +6% Fenhexamid 0.005 0.996 +3% Triclosan 0.001 0.994 +26% Simazine 0.001 0.999 -5% Acetochlor 0.01 0.997 0%

Metolachlor 0.001 0.995 -2%

ESI(+) Tetraconazole 0.01 1.000 +3%

-

LC Chlorfenvinphos 0.01 0.999 -2% Prochloraz 0.001 0.999 0% (Z)-Dimetomorph 0.005 0.999 +2%

Predicting the recovery through the physicochemical properties of the considered markers

Recovery can be measured using the standard addition method. Unfortunately, not all analytical standards are commercially available or accessible for cost or other practical reasons, especially

- 173/296 - Chapter 3. Development of non-targeted sample preparation protocols to screen halogenated markers of chemical exposure in human samples for new or unknown compounds sought by large-scale non-targeted screening approach. It is well-known that main factors influencing the recovery of a given compound are related to its physicochemical properties. Orthogonal-Partial Least Squares (OPLS) approach (Wold et al., 1989) is a known statistical tool to build regression models relating certain fixed parameters (named regressors, the X component of the model) involved in the prediction of one or more variables (named regressands, the Y component of the model). The main advantage of OPLS is that it is not negatively influenced by correlated variables. It can also handle missing values and vast datasets. A multivariate regression model based on OPLS approach was adopted to better characterise method limitations and efficiency. It was built using SIMCA-P 13.0 (Umetrics, Umea, Sweden). It evaluates whether the recovery of the compounds (matrix Y) could be predicted through their chemical structures (matrix X). The model was built with the 30 reference standard compounds characterised by 23 different physicochemical parameters extracted from PubChem, including: exact mass, mass defect, relative mass defect (RMD, Sleno, 2012), topological polar surface area, log P, complexity, number of each atom (C, H, N, O, Br, Cl, P, F, S), number of unsaturations, presence of OH group or oxygen double bond, presence of OH group, non-ramified cycle, heavy atom count, hydrogen bond donor count, hydrogen bond acceptor count, rotatable bond count (Appendix 8). Compound structure descriptors were regressed versus the measured recovery from LC or GC data, or the sum of both for acetochlor and metolachlor, for the concentration level 62.5 pg µL-1 of human milk. Due to the possible synergic effect, the squared and crossed terms of the descriptors were also included in the model. The obtained model was validated using k-fold cross-validation, and considered acceptable when R2 and Q2 indicators were above 0.5, with an associated p-value below 0.05. The model accuracy was estimated as root mean squared error (RMSE) (Eriksson et al., 2008).

The R2 and Q2 coefficients of the model were equal to 0.77 and 0.50 (respectively) with a p- value below 0.05 (0.0052), indicating a significant reliability. The model showed to be able to predict with good accuracy the recovery observed for our different reference compounds, with a RMSE and a cross-validated RMSE (RMSEcv) of 0.087 and 0.118 respectively. Predicted versus experimentally observed recovery for 30 test compounds used as training set (blue dots) in the OPLS model are reported in Figure 3.25. As shown, only three compounds (10%) are outside the RMSEcv limits. This means that the model is accurate in 90% of the cases, whatever the structural variability of the 30 compounds in the training set.

- 174/296 - Chapter 3. Development of non-targeted sample preparation protocols to screen halogenated markers of chemical exposure in human samples

Therefore, the regression model built in this study can be used to predict the recovery of compounds which are not available as chemical standards, using the structural descriptors retrieved from PubChem as input data. To corroborate this finding and evaluate over-fitting, the model was tested against a test-set of three compounds: atrazine, PFOS and BDCIPP. Their recovery was predicted using the model and it is plotted in Figure 3.25 (red dots). As shown, the error of prediction for the three compounds was of 0.106, 0.032 and 0.038, respectively. This result confirmed the accuracy of the OPLS model to predict compound recovery in the developed analytical method, with a relatively small error, based only on their chemical structures and associated physicochemical properties.

Figure 3.25: Results of the OPLS model built to predict recovery from a set of physicochemical properties of the considered biomarkers of exposure. Compounds 31, 32 and 33 were test compounds to assess model accuracy. Equation of the linear regression curve is Y = 0.9942x + 0.0006; R2 = 0.77; RMSEmin = 0.087 and RMSEmax = 0.118 are represented as upper and lower limits in black dotted line and grey dash line, respectively.

- 175/296 - Chapter 3. Development of non-targeted sample preparation protocols to screen halogenated markers of chemical exposure in human samples

The results of the OPLS model also allow to determine which parameters influence the most the recovery of the compounds. Taking into account the limited number of compounds of the model (n=30), we report here a list of variables that may contribute to the predictive model (VIP > 1 and correlation coefficient absolute value higher than 0.015) (Table 3.6).

Four main observations could be drawn: (1) Recovery increases for unbranched cycle molecule when the mass, complexity or number of heavy atom increases. (2) Higher is the molecular mass (>400-500 Da) or the log P (log P >5-6), lower is the recovery, except if the molecule contains OH group or oxygen double bond. (3) Higher is the molecular mass and/or the log P of a molecule containing heavy atoms, such as bromine, lower is the recovery. (4) It follows from these two last observations that higher is the degree of bromination, lower is the recovery except if the chemical structure contains OH group or oxygen double bond.

According to these observations, the method is evidencing a limitation for high molecular mass and/or hydrophobic molecules, especially when containing bromine atoms, except if the molecule contains OH group or oxygen double bond. This is in agreement with the fact that Captiva EMR-Lipid® sorbent selectivity is based on size exclusion and hydrophobic interaction. It could explain low recoveries for PBDE 153, anti-DP and BTBPE. However, low recoveries were also observed for smaller and more polar molecules as HCB and p-TBX. Signal of those molecules also decrease of more than 30% according to the matrix effect calculation. We hypothesised that those compounds can be well detected with the present analytical conditions in a clean system. However, the presence of matrix could impair their detection and this phenomenon is increased with high injector temperature and a long transfer line (more > 40 cm) which are parts of the system known for compound degradation.

Overall, these results highlight the complexity of non-targeted method development to combine a non-selective sample preparation with analytical system detecting a wide range of molecules. Thanks to the present OPLS model, another NTS method with complementary performance can be developed, with a specific focus on GC detection.

- 176/296 - Chapter 3. Development of non-targeted sample preparation protocols to screen halogenated markers of chemical exposure in human samples

Table 3.6: OPLS model observations with examples on related data. VIP is variable importance in projection.

Markers*markers (model component) VIP Correlation coefficient Reference in the text Example Complexity*Unbranched cycle 2.28 0.048 2,4-DCP (ND) and 2,4-DBP (ND) vs Heavy Atom Count*Unbranched cycle 2.18 0.043 (1) 2,3,4,6-tetra-CP (~20%) and 2,3,4,6-tetra- Exact Mass*Unbranched cycle 1.97 0.035 BP (~30%) Unbranched cycle 2.04 -0.048 XLogP3*OH group 2.24 0.045 Exact Mass*OH group 1.92 0.037 Fipronil Exact Mass*OH group or oxygen double bond 1.92 0.033 (2) OH-BDE 137 XLogP3 1.12 -0.017 TBBPA Exact Mass 1.17 -0.022 XLogP3*Exact Mass 2.10 -0.037 XLogP3*Heavy Atom Count 1.99 -0.025 BTBPE Exact Mass*Heavy Atom Count 1.99 -0.026 (3) p-TBX Exact Mass*Br 1.07 -0.029 PBDE 153 Exact Mass*Exact Mass 2.19 -0.040 Br*OH group 1.42 0.031 OH-BDE 137 Br*OH group or oxygen double bond 1.29 0.031 (4) TBBPA Hydrogen Bond Donor Count*Br 1.16 0.029 2,3,4,6-tetra-BP Br*Br 1.40 -0.036

- 177/296 - Chapter 3. Development of non-targeted sample preparation protocols to screen halogenated markers of chemical exposure in human samples

Comparison with targeted method

As previously detailed in the section 3.3.2.3, the non-targeted method can be assessed by comparison with targeted characterisation. The same targeted protocol dedicated to novel brominated flame retardants previously applied to characterise placenta samples was used to characterise the level of contamination in human milk. For this purpose, three human milk samples from a French mother-child study described elsewhere (Alexandre-Gouabau et al., 2019; Cano-Sancho et al., 2020), collected in 2008 and stored at -20 °C until analysis were analysed.

The same method used for placenta sample to calculate the minimal concentration detectable by the non-targeted method was applied to human milk prepared by the Captiva EMR-Lipid® protocol. In the present case, 400 µL of human milk were analysed and considering an average of 3 % of fat in milk and an estimated recovery of 30 % combined with the lowest instrumental LOD 1 pg µL-1, the lowest contamination level with our non-targeted method should be around 13 ng g-1 fat.

Results of targeted analysis (Appendix 6) showed a range of contamination levels from lower than the LOD to 6 ng g-1 fat, with most values around 0.5 ng g-1 fat (Figure 3.26). In the best case scenario, i.e. without any loss during sample preparation and analysis, only samples with the highest level of contamination could be then characterised with the non-targeted approach. However, even if the limit of detection of the non-targeted method is insufficiently low in the current state to detect this list of nBFR in those human milks, it is in adequacy with contamination levels in human samples for highly exposed individuals.

- 178/296 - Chapter 3. Development of non-targeted sample preparation protocols to screen halogenated markers of chemical exposure in human samples

Figure 3.26: Concentration (ng g -1 lipid) of organic contaminants detected in human milk samples prepared with the targeted method developed by LABERCA and analysed in GC-EI-magnetic sector, LC-ESI-QqQ, LC-ESI-Orbitrap, APGC-QqQ.

3.4.4. Compatibility with EDA approach

Partnership context

In the frame of a collaboration with WP16 partners of the Environment and Health department of Vrije Universiteit Amsterdam (VU-E&H), a complementary approach based on biological activity was investigated. This collaboration was focused on endocrine disruptors with a biological test focused on activity mediated through a thyroid hormone receptor interaction. The approach, effect-direct analysis (EDA), combines a sample fractionation in LC with a dual

- 179/296 - Chapter 3. Development of non-targeted sample preparation protocols to screen halogenated markers of chemical exposure in human samples detection between mass spectrometry for chemical (structural) information and fluorescence for biological responses (more information are detailed in Chapter 1). By correspondence between both detection modes, compounds disrupting the thyroid hormone system are highlighted by fluorescence and identified by MS.

The purpose of this collaboration was to extend the applicability of the Captiva EMR-Lipid® sample preparation to another approach (biological approach), which could reinforce its adequacy with non-targeted screening assessing, thus the biological approach to characterise human exposure without a priori knowledge.

The analysis of the whole extract allows to estimate the mixture effect of contaminants of the organism. Both EDA and whole extract analysis approaches were tested on human milk previously prepared with the Captiva EMR-Lipid® protocol.

EDA protocol

A human milk sample (400 µL) was prepared with the Captiva EMR-Lipid® protocol detailed earlier (Appendix 7). The compatibility of the Captiva EMR-Lipid® sample preparation with EDA was assessed with a mix of contaminants (Table 3.7) proposed by VU and known for their interaction with TTR (i.e. properties to disrupt thyroid hormone system by competition with T4 to be bound to TTR). Two levels of concentration were used by spiking 1 and 4 µL of mix, in addition, a sample and a procedural blank were not spiked and used as control. After the liquid- liquid partitioning, the polar fraction was split into two equal fractions, reconstituted in i) 200 µL DMSO/water 4:1 (v/v) for the whole extract bioassay and ii) in 200 µL water/methanol 4:1 (v/v) for the EDA (LC analysis and fractionation). Protocols for bioassays, fractionation and LC-HRMS analysis were developed by VU (Ouyang et.al., 2017) and details are in Appendix 9.

- 180/296 - Chapter 3. Development of non-targeted sample preparation protocols to screen halogenated markers of chemical exposure in human samples

Table 3.7: List of standards (thyroid hormone system disruptors) proposed by VU to spike samples for EDA with the concentration in the mix in ng µL-1.

Name Acronym Concentration Tetrabromobisphenol A TBBPA 174 2,4,6-Tribromophenol 2,4,6-TBP 83 5-hydroxy bromodiphenyl ether 47 5-OHBDE47 75 6-hydroxy bromodiphenyl ether 47 6-OHBDE47 502 6-hydroxy bromodiphenyl ether 99 6-OHBDE99 232 4-hydroxy chlorobiphenyl 107 4-OHCB107 205 4-hydroxy chlorobiphenyl 187 4-OHCB187 62

Whole extract bioassay analysis

EDA is a qualitative approach that is complementary of the quantitative approach determining the amount of TTR-binding activity in the whole extract sample. The whole extract approach is used in toxicology in order to establish a relationship between the concentration of toxic substances (thyroid hormone system disruptors) and the observed effect (biological response). Contaminants of interest are not identified with this approach but the mixture effect (i.e. the whole effect of contaminants on the thyroid hormone system) can be estimated. Furthermore, this approach could be used to prioritise samples regarding their activity against a specific receptor, indicating certain level of contamination. This is interesting especially for NTS in human biomonitoring studies, in order to first focus on highly contaminated samples and increase the probability to detect emerging compounds.

For the whole extract bioassay, each sample was diluted by 1, 3, 10 and 30 and incubated in identical reaction mediums (same concentrations of TTR and T4-FITC) and the fluorescence response corresponds to the percentage of T4-FITC-TTR binding. At 20 % of fluorescence T4- FITC is not bound to the TTR because of a competition with toxicants or natural T4 remaining in the extract. At 100 % all T4-FITC are bound to the TTR and no toxic effect is observed. The concentration of toxicants is calculated on the dilution curve at 20 % of T4-FITC signal inhibition, i.e. 80 % of fluorescence. For the present study, the toxicological approach and concentration estimation was not further investigated. The whole extract bioassay was used to visualise the global thyroid hormone system disruption and the compatibility of this approach

- 181/296 - Chapter 3. Development of non-targeted sample preparation protocols to screen halogenated markers of chemical exposure in human samples to prioritise highly contaminated samples. As illustrated in Figure 3.27, fluorescence signal for samples spiked with the VU mix is less intense than for unspiked sample and procedural blank. This T4-FITC signal inhibition demonstrated the toxic effect of those compounds on the thyroid hormone system. Moreover, it proves the compatibility of the approach to prioritise highly contaminated samples. In the same figure, the fluorescence responses of samples spiked with standards of the QA/QC mix 2, i.e. concentration in adequacy with human exposure, are similar to procedural blank. This highlights the lack of sensitivity of the bioassay to capture the low environmental levels of human internal exposure.

Different perspectives have been identified to improve the sensibility of the bioassay either for EDA or whole extract analysis. Beginning with a larger sample volume would concentrate compounds of interest to reach lower LOD. The extract could be reconstituted in a lower volume to concentrate the sample. The addition of buffer, TTR, T4-FITC for the bioassay induces 200-fold sample dilution. This protocol is under improvement by WP16 partners to reduce this dilution.

- 182/296 - Chapter 3. Development of non-targeted sample preparation protocols to screen halogenated markers of chemical exposure in human samples

Figure 3.27: Fluorescence response of whole extract bioassay of human milk prepared with the Captiva EMR-Lipid® protocol and spiked with 1 µL (blue triangle) and 4 µL (red triangle) of the VU mix and compared to procedural blank (blue dot) and unspiked milk (black rectangle). The same sample was spiked with 50 µL (orange diamond), 200 µL (grey diamond) and 400 µL (green diamond) of standards of the QA/QC mix 2.

- 183/296 - Chapter 3. Development of non-targeted sample preparation protocols to screen halogenated markers of chemical exposure in human samples

Compatibility of the Captiva EMR-Lipid® sample preparation with EDA

After fractionation of each sample (spiked and unspiked), the biological activity was measured in each extract. Results of these EDA measurements are illustrated in Figure 3.28. A biological activity (i.e. interaction of thyroid hormone system disruptor with TTR) is characterised by a decrease in fluorescence signal. This is observed for spiked samples where some of the added compounds interact with the TTR system. As expected, the measured biological activity increases with the concentration level of the fortified samples. In parallel of the fluorescence response, the added standards were detected by LC-ESI-HRMS. The fluorescence response was proven to be linked to these spiked standards and not to other compounds ‘naturally’ present in the sample which could co-eluate at this retention time, since no biological activity was observed in the unspiked sample. Spiked standards have hazard effect on the thyroid hormone system, as was expected based on VU expertise. Therefore, the detection of these standards by LC-HRMS and fluorescence proved the compatibility of the sample preparation with the EDA approach to detect thyroid hormone system disruptors, at least for relatively high concentrations.

Figure 3.28: Results of the performed EDA assay with the biological response detected by fluorescence (top) for spiked (1 µL in blue and 4 µL in red) and unspiked milk (grey) overlapped with the corresponding extracted ion chromatograms obtained for standards detected by LC-ESI(-)-HRMS.

- 184/296 - Chapter 3. Development of non-targeted sample preparation protocols to screen halogenated markers of chemical exposure in human samples

This first experiment ran with VU standards at high concentrations (> 60 ng spiked in sample) was repeated with the standards of the QA/QC mix 2. The same human milk was spiked with 5, 20 and 40 ng of standards and analysed in the same conditions. In comparison, all other experiments described in the present manuscript were assessed with 5 ng of standards spiked in samples. As illustrated in Figure 3.29, no biological activity was recorded for the three spiking levels whereas standards were well detected by LC-HRMS. According to the previous experiment where standards at higher concentration were detected, the present result highlights a lack of sensitivity of the bioassay, thus impeding its compatibility with environmental concentration levels of human chemical exposure.

- 185/296 - Chapter 3. Development of non-targeted sample preparation protocols to screen halogenated markers of chemical exposure in human samples

Figure 3.29: Results of the performed EDA assay with the biological response detected by fluorescence (top) for human milk spiked with 50 µL (top), 200 µL (middle) and 400 µL (bottom) standards of the QA/QC mix 2 overlapped with the corresponding extracted ion chromatograms obtained for standards detected by LC-ESI(-)-HRMS.

- 186/296 - Chapter 3. Development of non-targeted sample preparation protocols to screen halogenated markers of chemical exposure in human samples

Beyond this global assessment, the bioassay sensitivity was more particularly assessed for TBBPA. To summary, samples were spiked with TBBPA at 0.01; 0.05; 0.1; 0.4; 1.7 ng µL-1 milk. According to the previous experiment, biological activity was detected for milk sample extracts fortified at 0.4 and 1.7 ng µL-1 but was not detected at 0.1 ng µL-1. Thus, the LOD of this bioassay for TBBPA in human milk prepared with the Captiva EMR-Lipid® protocol was estimated at 0.4 ng µL-1 milk. As a comparison, Inthavong et al. (2017) published a comparison between various TBBPA exposure levels detected in human milk, reported from several studies. The upper level of contamination to TBBPA was 688 ng g-1 lipid. Considering that human milk contains 3 % of lipids and 1 mL milk is equivalent to 0.03 g lipids, TBBPA was detected at 0.02 ng µL-1 milk. The bioassay LOD then appears 20 times higher than the expected upper level of naturally occurring human exposure level. To conclude, this approach is very promising within a non-targeted context but the current lack of sensitivity is impairing its applicability for capturing the low environmental levels of human internal exposure. The case of TBBPA illustrates the difference of sensitivity between mass spectrometry and bioassay but should not be viewed as a generality since the LOD depends on each compound.

3.4.5. Discussion and conclusion

The last sample preparation development was conducted thanks to conclusions learnt after the assessment of other approaches (acidic hydrolysis, GPC, Bligh and Dyer). The present protocol applied to human milk and meconium removes lipids with the Captiva EMR-lipid® stationary phase. Then, the liquid-liquid partitioning allows to split interferences into two fractions to reduce matrix effect and promote LC- and GC-HMRS complementarity.

Classical criteria such as linearity, recovery and matrix effect were assessed and were judged acceptable. The sensitivity of present method is in adequacy with concentration ranges detected in human samples. Nevertheless, it is assumed that the sensitivity of non-targeted methods is not as low as targeted methods, which illustrates the complementarity of both. It also demonstrates challenges regarding method development and that non-targeted approaches should not be developed in the scope of replacing targeted methods.

Some methods limitations and strengths were drawn in terms of method application range with the QA/QC mix 2, but it does not deal with other molecules and especially with unknowns. The multivariate regression model constructed on the results of the detection of standards spiked in samples enables to predict with good accuracy the recovery of novel identified compounds. In

- 187/296 - Chapter 3. Development of non-targeted sample preparation protocols to screen halogenated markers of chemical exposure in human samples general, the model indicated that bigger and less polar molecules lacking alcoholic or ketone groups will be less accessible through the developed method. Thanks to the results of the prediction model, it is assumed that other methods more focused on heavy and brominated compounds are required to complete the capabilities of the present method.

The present sample preparation was also judged compatible with EDA approach in case of higher exposure level. However, the sensitivity of the bioassay for thyroid hormone system disruptor detection remains limited for capturing the low environmental levels of human internal exposure, but the sample preparation was demonstrated to be compliant and compatible with this biological approach. Therefore, this sample preparation can be extended to chemical and biological characterisation of human samples. In addition, as lipids are retained on the Captiva EMR-lipid® stationary phase and as this phase is desegregated with dichloromethane, lipids could be released and analysed in a context of lipidomic studies.

- 188/296 - Chapter 3. Development of non-targeted sample preparation protocols to screen halogenated markers of chemical exposure in human samples

3.5. Conclusion

This third chapter was dedicated to the investigation of several sample preparations to purify various human matrices in the scope of NTS characterisation. The combination of all observed issues and challenges allowed us to draw first guidelines for laboratories wishing to start with non-targeted screening of human matrices. The present Ph.D. project put the first stone of the edifice, and interesting perspectives of improvement were highlighted.

We first propose that a non-targeted sample preparation should be developed in regards of the main matrix interferences (for the present study lipids and proteins) that should be removed. For instance, protein precipitation with organic solvent and lipid removal by chemical reaction or size. In addition, a compromise between selectivity and sensitivity has to be found to remove matrix interferences while preserving contaminants of interest. However, the lack of purification can damage the signal detection by matrix effect and carry-over. Thus, sample extracts need a minimal level of compatibility with analytical instrumentations. This can be adjusted with different initial sample amounts to reach sufficient sensitivity while promoting purified extract.

Then, our first advice is to evaluate the sample amount to be considered for analysis in function of biological accessibility and sample preparation purification level. The efficiency of the purification should include an estimation of the matrix effect and especially ion suppression phenomenon.

The second advice is to constitute a mix of standard representative of a wide range of molecules, thus supposed to constitute the application range of the developed method. It can be used to assess method strengths and limitations. It appears very promising when results related to this mix of standard are integrated in a predictive model. The ideal case would be to combine different mixes of standards to enrich the model with additional result and increase its predictive accuracy, step-by-step.

Last but not least, the main question to characterise a sample with NTS approach is “if a compound is not detected, is it because the sample was not contaminated or because the analytical method was not efficient enough to detect it?”. The complementarity of targeted and non-targeted methods is well illustrated in this question and the dual characterisation of samples by those two approaches should be promoted.

- 189/296 - Chapter 3. Development of non-targeted sample preparation protocols to screen halogenated markers of chemical exposure in human samples

Finally, the management of the external contamination should be considered since the beginning of the sample preparation development. Bioinformatics tools could deal with this issue but, just like matrix effect, it could suppress signals of interest. Moreover, it is a source of false-positive results which should be avoided.

To conclude, the development of four sample preparation protocols on four human matrices was investigated and allowed us to list these guidelines. The comparison of these approaches is summarised in Table 3.8. As strength and limitations were assessed for those methods, they can be applied to real sample analysis. This is presented in Chapter 4.

Table 3.8: Summary of developed sample preparation for the non-targeted characterisation of human samples, including comparison of their capabilities.

Acidic Captiva EMR- GPC Bligh and Dyer hydrolysis Lipid® Ease of use + + + +

Throughput + - + + Green +/- - + + chemistry Matrix Proteins Proteins Proteins interference Proteins Lipids Lipids Lipids removal Matrix of High lipid Solubilised in Solubilised in In theory “all” interest content water water Human milk Example Adipose tissue Adipose tissue Placenta Meconium

- 190/296 -

4. CHAPTER 4

CHARACTERISATION OF INTERNAL HUMAN EXPOSURE TO HALOGENATED CONTAMINANTS IN HUMAN SAMPLES BY NON-TARGETED SCREENING

- 191/296 -

Chapter 4. Characterisation of internal human exposure to halogenated contaminants in human samples by non- targeted screening

4.1. Introduction

The present chapter deals with several concrete applications of the developed non-targeted screening strategies to analyse various types of human samples and thus reveal halogenated markers of internal exposure. The different sample preparation protocols and analytical methods presented in Chapter 2 and 3 were applied to characterise the chemical exposure during the perinatal period by analysing adipose tissue, placenta, human milk and meconium samples. These applications, respectively described in the following sections, illustrate the faced challenges and highlight future work of interest for the development of non-targeted approaches.

Adipose tissue appears as an interesting matrix to seek POPs, since it is a compartment of storage for many lipophilic compounds (high log P) (Li et al., 2006; Mustieles et al., 2017). Even if these samples are only accessible by surgical intervention, they can be accessible through several surgeries (e.g. caesarean, liposuction).

Placenta is a non-invasive matrix and it represents the link between the mother and the foetus during the pregnancy and some contaminants can either diffuse or being transported through the placenta to the foetus (Myren et al., 2007; Gützkow et al., 2012).

Human milk is also a non-invasive matrix and it illustrates the link between the mother and the newborn after delivery and up to six months (Nickerson, 2006). It allows to characterise the maternal (past exposure of the mother) and newborn exposure (route of exposure by food).

Meconium is the first baby stool after delivery. It is accumulated from the third-fourth of gestation and released within several hours up to three days after the birth (Ortega García et al., 2006). As opposed to newborn urine and blood/cord blood, meconium is a non-invasive matrix for characterising perinatal exposure.

4.2. Analysis of adipose tissue samples

4.2.1. Introduction

Adipose tissue was considered in the present Ph.D. work as a first proof-of-concept matrix, to identify the main challenges that have to be faced in order to characterise human sample with

- 193/296 - Chapter 4. Characterisation of internal human exposure to halogenated contaminants in human samples by non- targeted screening a non-targeted approach, considering the known bioaccumulation of halogenated lipophilic markers of exposure in this particular biological storage compartment.

Human adipose tissue and procedural blank were prepared with the acidic hydrolysis protocol detailed in Chapter 3 (section 3.2.2.1 Sample preparation and analysis). Mussel sample was also prepared, in a context of another Ph.D. work occurring at LABERCA, to be used as a comparison with human adipose tissue.

4.2.2. Adipose tissue sample investigation

The investigation and comparison of MD plots obtained from HaloSeeker v1.0, after analysis of procedural blank, human adipose tissue and mussel samples (Figure 4.1) led us to several observations.

Spiked standards of 13C-TBBPA, 2H-α-HBCDD and 2H-γ-HBCDD were easily identified on the MD plot obtained from human adipose tissue. Moreover, as illustrated in the plot, HBCDD adducts can be clearly distinguished. This confirmed the efficiency of the software and its compatibility to process HRMS data. This observation also enabled to guide and prioritise further implementation of additional functionalities in the software. For instance, the CAMERA package for aggregating the multiple adducts species belonging to a same compound is now embedded in HaloSeeker v2.0.

The difference of information observed on the MD plot resulting from the procedural blank and the adipose tissue samples then did not appeared immediately visible. Thus, even if this software enables to visually prioritise signals of interest, the present example demonstrates the need of bioinformatics tools to detect and manage (e.g. subtract) signal detected in both procedural and matrix samples.

As a comparison, the MD plot obtained from a mussel sample, purified with exactly the same sample preparation and analysed with the same LC-ESI(-)-HRMS settings, exhibited a significantly higher number of clusters than procedural blank and human adipose tissue. Mussels clearly appeared much more contaminated than human adipose tissue. Several exposure markers from different substance classes, including chlorophenols, bromophenols and hydroxy-chloro/bromo-diphenyl ether were identified and are labelled in Figure 4.1. This observation is reflecting the difference in terms of level of contamination between biota and

- 194/296 - Chapter 4. Characterisation of internal human exposure to halogenated contaminants in human samples by non- targeted screening human samples, thus constituting an additional challenge for the latter when conducting non- targeted screening analyses.

Overall, this observation allowed to confirm the global suitability of the developed non-targeted approach, including the use of HaloSeeker to prioritise halogenated compounds. It also confirmed that the characterisation of the chemical exposure in human matrix is more complex than the analysis of most environmental and food matrices, especially due to the more restricted panel and lower abundance of the accessible markers of exposure.

The generated data were processed with the strategy of annotation described in Chapter 2 (2.4.2 Strategy developed to process data generated by non-targeted approach). Hexachlorocyclohexane (HCH) was an example of marker annotated in human adipose tissue - through the adduct [M+CH3CO2] thanks to the annotation reference database elaborated in the frame of the HBM4EU project. The retention time was verified by the injection of pure standard of β-HCH. MS/MS experiments were conducted at NCE 10, 30 and 60% but the signal abundance was insufficient. The resulting fragmentation spectra did not lead to the validation of the compounds identity.

Another compound, α-HBCDD, was identified at the same retention time as the corresponding added labelled standard (2H-α-HBCDD). The fragmentation of HBCDD is characterised by a single fragment of 79/81Br, which was observed for both pure analytical standard and the compound detected in the matrix. Thus, β-HCH was presumably identified and α-HBCDD was identified in human adipose tissue.

This result demonstrates the capability of the developed approach to reveal such halogenated markers of human internal chemical exposure without a priori knowledge. The detection of a compound such as β-HCH also confirms a previous hypothesis that this extract should benefit from GC analysis in complement to LC, as these lipophilic compounds are better detected in GC. From this preliminary encouraging result, additional time should be allocated to this data processing and analysis, especially with further improvement of the annotation and identification strategy, in order to reveal and possibly identify more markers of exposure from the same sample.

- 195/296 - Chapter 4. Characterisation of internal human exposure to halogenated contaminants in human samples by non- targeted screening

Figure 4.1: MD plot obtained with HaloSeeker v1.0 software following the data processing of a procedural blank (top), 1 g human adipose tissue (middle) and 0.5 g fat from mussel (bottom) samples purified by acidic hydrolysis and analysed by LC-ESI(-)-HRMS under the same conditions.

- 196/296 - Chapter 4. Characterisation of internal human exposure to halogenated contaminants in human samples by non- targeted screening

4.2.3. Conclusion

The investigation of real samples allowed to validate a first proof-of-concept. However, the high complexity of this matrix and some difficulties encountered while dealing with this fatty material led us to deprioritise it relatively rapidly in favour of other human matrices. The extent of results obtained on adipose tissue therefore appears, in its current state, quite limited.

The obtained results then illustrate how challenging the development of non-targeted approach from the sample preparation to data processing can be. Moreover, lipophilic compounds were expected and these analytes, such as PCB, are normally detected in GC rather than by LC. It shows the importance of the LC and GC complementarity to perform comprehensive analysis. At the time of these experiments, such GC-HRMS instrument was not yet installed in the laboratory (this is the case now) and a perspective of this first human matrix investigation will be to characterise our extracts prepared from human adipose tissue by GC-HRMS in order to reveal more markers of internal exposure than those presently identified at this stage.

4.3. Analysis of placenta samples

4.3.1. WP14-WP16 interaction

In the frame of the HBM4EU project, a collaboration between WP14 and WP16 partners was initiated in order to characterise the human foetal exposure through a combination of chemical and biological profiling approaches applied to placenta samples. Samples originating from the INMA (Environment and Childhood) cohort in Spain (Ribas-Fitó et al. 2006, Guxens et al., 2011) were analysed by WP14 partners with different bioassays in order to reveal markers of exposure by biological approaches. (HBM4EU, Deliverable report AD14.4, 2019). In parallel and at French level, WP16 partners (INRAE, INSERM and CEA) worked together to complete the multi-characterisation of the same samples with chemical approaches. Then, correlations between chemical and biological approaches to establish the link between markers of exposure and their health effects were assessed. In a first step, and separately from the present Ph.D. work, chemical approaches based on targeted analysis of PFAS and steroid hormones were conducted. In a second step, in the context of the present Ph.D. work, the non-targeted characterisation of those placenta samples was performed.

- 197/296 - Chapter 4. Characterisation of internal human exposure to halogenated contaminants in human samples by non- targeted screening

The following section deals with this non-targeted analysis, conducted during the present Ph.D on 25 placenta samples. project. It is based on the previously detailed Bligh and Dyer sample preparation (Chapter 3, section 3.3.2.1 Protocol) and analytical strategy (Chapter 2, section 2.2.2.1 LC method development) that we developed. In addition to placenta samples, three QC (pool of the 25 placentas) and six procedural blanks were analysed in a single batch. Blank and QC were used to control the external contamination and to validate the analytical batch. According to the previous results on the partitioning of standards and the chromatogram profile (Chapter 3, section 3.3.2.2 Method performance), only chloroform fractions were analysed and investigated.

4.3.2. Placenta sample investigation

All prepared placenta samples and procedural blanks were analysed by LC-ESI(-)-HRMS and an example of chromatograms obtained for these extracts is provided in Figure 4.2.

Figure 4.2: Overlap of total ion currents obtained for the chloroform fraction of placenta (blue) and procedural blank (black) samples purified with a Bligh and Dyer extraction and analysed by LC-ESI(-)-HRMS.

LC-ESI(-)-HRMS data generated for the chloroform fraction were processed with HaloSeeker v1.0 and signals of interest were prioritised. In total, 848 clusters (signal detected for a group of halogenated ions) were aligned for 25 placentas and six procedural blanks. This high number of clusters was highlighted by considering the frequency of detection of signals though the

- 198/296 - Chapter 4. Characterisation of internal human exposure to halogenated contaminants in human samples by non- targeted screening number of samples. Clusters detected in at least 10/25 placentas were first investigated. After blank subtraction and manual investigation, 69 clusters were highlighted and 12 clusters were putatively annotated with the HBM4EU annotation reference database. However, in those 12 hits, no successful unambiguous identification was obtained. Then, other clusters less frequently detected (i.e. in 1 to 9 placentas out of 25) were investigated with the same methodology. During this second round of analysis, 113 clusters were highlighted and 12 were putatively annotated with the HBM4EU list. In particular, a cluster detected in three placentas was annotated as triclosan (Figure 4.3). Mass spectra (experimental and theoretical) for the three placentas matched at 49, 57 and 83 % (score obtained from HaloSeeker v1.0) with -0.09; -0.02 and -0.01 ppm of mass deviation, respectively.

Figure 4.3: MD plot obtained from HaloSeeker v1.0 software for a placenta sample purified with a Bligh and Dyer extraction and analysed by LC-ESI(-)-HRMS, with a zoom on the cluster annotated as triclosan.

The pure analytical standard of triclosan was analysed in the same chromatographic conditions and was eluted at the same retention time (difference lower than 0.1 min) than the compound detected in placentas. Pure analytical standard of triclosan (0.5 ng loaded on column) was fragmented at NCE 10; 30 and 60%. Precursor ion (286.94388) was detected at NCE 10% and diagnostic ions (141.98264 and 160.95651 m/z) were detected at NCE 60% (Figure 4.4).

- 199/296 - Chapter 4. Characterisation of internal human exposure to halogenated contaminants in human samples by non- targeted screening

Figure 4.4: Fragmentation mass spectra of triclosan (precursor ion 286.94388 m/z) detected by LC-ESI(-)-HRMS for low (NCE 10%) and high (NCE 60%) collision energies (top and bottom, respectively), with the annotation of the observed diagnostic fragments.

The signal detected in placenta samples was fragmented at NCE 10 and 60% but neither precursor ion nor diagnostic ion was detected, probably because of the insufficient concentration of the compound in the extract. A concentration step in order to enhance this signal was unfortunately not feasible because of the low volume of extract available. The identity with a confidence level 1, according to the scale proposed by Schymanski et al., 2015, was then not confirmed but the annotation reached a high level of certainty thanks to the retention time and isotopic pattern.

- 200/296 - Chapter 4. Characterisation of internal human exposure to halogenated contaminants in human samples by non- targeted screening

A semi-quantitative estimation was performed based on the results of recovery assessed for triclosan in Chapter 3 (triclosan was included in the standard mix 1 of Cl/Br-phenolic compounds). According to this result, the concentration in triclosan was semi-quantitatively estimated between 0.7 and 1 ng g-1 placenta (70-100 ng g -1 fat) in the three samples. Targeted analysis developed for triclosan analysis is now required in order to generate accurate quantitative data.

4.3.3. Conclusion

Our developed non-targeted screening strategy was applied to a set of 25 placenta samples and the nonpolar fraction resulting from the Bligh and Dyer sample preparation applied to these samples was more specifically investigated. The vast amount of detected signals through placenta samples, QC and procedural blank was mainly prioritised according to a detection frequency based criteria. This approach is very interesting to compare two groups of population (e.g. case/control), but does not represent a perfect standard. Indeed, the chemical exposure levels and patterns observed in human are a matter of high inter-individual variability according to various factors, and compounds detected in only a particular sub-population may be of interest and concern in terms of risk assessment. At the present stage, the identification of compounds revealed by NTS is still laborious and the prioritisation of the detected signals using a detection frequency criterion is interesting as a first intention, but a broader range of detected compounds should be investigated nonetheless. In the present case, all detected signals were finally investigated and triclosan was identified in three placentas. This result was obtained after days of data processing and considerable manpower. The recent update of HaloSeeker v2.0 and other approaches such as iterative DDA should reduce this time. Then, signals generated by NTS will be deeper investigated and more successful annotations are expected.

4.4. Analysis of human milk samples

4.4.1. Samples and protocol

Human milk samples used for the present method development were originated from a French mother-child study described elsewhere (Alexandre-Gouabau et al., 2019; Cano-Sancho et al., 2020), collected in 2008 and stored at -20 °C until analysis. Five human milk were thawed, vigorously agitated and pooled (2.5 mL each) to constitute a composite sample further used for analysis.

- 201/296 - Chapter 4. Characterisation of internal human exposure to halogenated contaminants in human samples by non- targeted screening

The composite milk was prepared in triplicate with the Captiva EMR-Lipids® sample preparation described earlier (Appendix 7). Samples were analysed with LC- and GC-HRMS methods detailed in Chapter 2 (sections 2.2.2 LC-HRMS method development and 2.2.3 GC- HRMS method development) and data was processed with the strategy of identification detailed in the same chapter (section 2.4.2 Strategy developed to process data generated by non-targeted approach).

4.4.2. Human milk sample investigation

Investigation of LC-HRMS data

The polar fraction (acetonitrile/water fraction) resulting from the prepared human milk and procedural blank samples were analysed by LC-ESI(-)- and ESI(+)-HRMS and an example of chromatograms obtained for those extracts is provided in Figure 4.5.

- 202/296 - Chapter 4. Characterisation of internal human exposure to halogenated contaminants in human samples by non- targeted screening

Figure 4.5: Overlap of total ion currents obtained for the polar fraction of human milk (blue) and procedural blank (black) samples purified with the Captiva EMR-Lipid® protocol and analysed by LC-ESI(-)- and ESI(+)-HRMS, top and bottom respectively.

In total, 286 and 205 clusters were aligned in the triplicate composite milk samples analysed in LC-ESI(-)- and ESI(+)-HRMS, respectively, from which 45 and 12 clusters were highlighted after both blank subtraction and manual investigation. Then, based on those, 17 and 5 clusters matched with the HBM4EU annotation reference database, respectively. In the 17 hits detected in LC-ESI(-)-HRMS, an isotopic cluster matched with 4-hydroxychlorothalonil, a metabolite of chlorothalonil (pesticide). Isotopic mass spectra (experimental and theoretical) matched at 91% (score obtained from HaloSeeker v1.0) with 0.4 ppm of mass deviation (Figure 4.6-A2).

- 203/296 - Chapter 4. Characterisation of internal human exposure to halogenated contaminants in human samples by non- targeted screening

To confirm the identification, pure analytical standard of 4-hydroxychlorothalonil was analysed with the same analytical conditions. Both pure analytical standard and the compound detected in the milk sample were eluted at 4.67 and 4.63 min, respectively. Both were fragmented at NCE 60% and the following peak list (with mass deviation in ppm in brackets) illustrates the perfect match between the compounds detected in milk and the pure analytical standard of 4- hydroxychlorothalonil: 244.90799 (-0.7); 209.93935 (0.2); 181.94470 (1.6); 174.97049 (0.2) and 146.97565 (0.7) (Figure 4.6-B). According to the confidence level proposed by Schymanski et al., 2015, 4-hydroxychlorothalonil was then identified in human milk at confidence level 1, based on a non-targeted approach.

Figure 4.6: Identification of an unknown halogenated cluster detected human milk purified with the Captva EMR-Lipid® protocol and analysed by LC-ESI(-)-HRMS. (A1) Mass defect plot generated by HaloSeeker v1.0 with polyhalogenated clusters detected in unspiked samples only. The halogenated cluster at m/z 244 is zoomed in the rectangle. (A2) Experimental mass spectrum of this cluster and, in mirror, the theoretical mass spectrum of hydroxy-

chlorothalonil (C8HCl3N2O) proposed by HaloSeeker v1.0. (B) Fragmentation mass spectra at NCE 60% of the unknown cluster (top) and of the pure analytical standard of 4- hydroxychlorothalonil (bottom).

- 204/296 - Chapter 4. Characterisation of internal human exposure to halogenated contaminants in human samples by non- targeted screening

The recently developed iterative DDA (iDDA) approach (Chapter 2, section 2.4.2.1 Investigation of LC-HRMS data) was applied on this composite of human milk to generate complementary MS/MS data without a priori knowledge. It reinforces structural identification score and further discovery of markers of exposure through another independent strategy. The 4-hydroxychlorothalonil was fragmented at NCE 10, 30 and 60%. Precursor and diagnostic ions were detected with a mass accuracy lower than 1 ppm, which validate the identity of the 4- hydroxychlorothalonil with the iDDA approach (Figure 4.7).

Figure 4.7: Extracted ion chromatogram of 4-hydroxychlorothalonil detected in human milk samples purified with the Captiva EMR-Lipid® protocol and analysed by LC-ESI(-)-HRMS (left top) with the corresponding three scans for the fragmentation of the precursor ion (m/z 244.98) with the iDDA approach (left bottom), and the resulting spectra of fragmentation (NCE 10, 30 and 60%) (right).

To the best of our knowledge, this is the first time that 4-hydroxychlorothalonil was identified in human milk. However, chlorothalonil is known in analytical chemistry field to face difficulties regarding stability during extraction and injection. As an example, the labelled analytical standard of chlorothalonil was spiked in human milk before extraction and only 4- hydroxychlorothalonil was detected. Also, the pure analytical standard of chlorothalonil injection resulted in two peaks at different retention time both with the same mass spectrum of 4-hydroxychlorothalonil. It is known that chlorothalonil is easily transformed into its

- 205/296 - Chapter 4. Characterisation of internal human exposure to halogenated contaminants in human samples by non- targeted screening hydroxylated metabolites during sample preparation and/or injection (Kiefer et al., 2019; EURL report “Modified QuEChERS-Method for the Analysis of Chlorothalonil in Fruits and Vegetables”).

Moreover, this compound was not detected in procedural blank samples, but a potential presence originating from upstream sample collection and/or field work was not possible to investigate in the present case, so the exact origin of this observed contamination is still somewhat unsure. A publication from LABERCA dealing with this topic is currently under review. This observation highlights the complexity and strength of NTS approach as it allows to detect a wide range of chemical contaminants independently of the origin.

Consequently, if milk was externally contaminated by chlorothalonil during sample collection (for instance with containers), its metabolite could be detected in the final extract. Thus for the present study, it is not possible to determine if the molecule detected in the human milk, and so potentially ingested by the baby, was chlorothalonil or its hydroxylated metabolites or both. According to this state of uncertainty, it is not directly possible to conclude that the breastfed baby was exposed 4-hydroxychlorothalonil. However, the non-targeted approach allowed us to confirm the presence of an organic contaminant, detected without a priori knowledge, and raise questions either regarding the exposure of the mother/newborn or the ubiquitous nature of this compound.

Investigation of GC-HRMS data

The nonpolar (hexane) fraction resulting from the developed sample preparation applied to the same human milk and procedural blanks were also analysed by GC-EI-HRMS and an example of chromatograms obtained for those extracts is provided in Figure 4.8.

- 206/296 - Chapter 4. Characterisation of internal human exposure to halogenated contaminants in human samples by non- targeted screening

Figure 4.8: Overlap of total ion currents obtained for the nonpolar fraction of human milk (blue) and procedural blank (black) samples purified with the Captiva EMR-Lipid® protocol and analysed by GC-EI-HRMS.

From the generated GC-HRMS data, HCB and p,p’-DDE were identified at confidence level 1, according to the Schymanski scale (Figure 4.9). HCB and p,p’-DDE were detected at concentration estimated to 4 and 2 ng mL-1. Those two historical chemical contaminants have already been detected in human milk in several studies (Campoy et al., 2001; Fürst, 2006; Aerts et al., 2019) and they validate the efficiency of the present method and its compatibility for detecting environmental contaminants. However, the developed method is facing some limitations regarding the nature and concentration of accessible markers. In particular, the identification step in GC-HRMS requires advanced software addressing halogenated pattern issue and real database with experimental and/or in-silico fragmentation spectra.

- 207/296 - Chapter 4. Characterisation of internal human exposure to halogenated contaminants in human samples by non- targeted screening

Figure 4.9: Identification of p,p'-DDE (A) and HCB (B) in human milk (top) by GC-EI-HRMS with comparison to spectra fragmentation of analytical standard (bottom). Peak list with mass deviation in ppm in brackets for p,p’-DDE: 317.93441 (0.03); 280.96852 (0.5); 245.99962 (0.3); 210.02285 (0.5) and for HCB: 283.80967 (-1.0); 248.84095 (-0.1).

4.4.3. Conclusion

Human milk samples were characterised with a non-targeted approach including a comprehensive analysis based on a combined LC- and GC-HRMS detection. The LC and GC complementarity allowed to detect and identify 4-hydroxychlorothalonil and p,p’-DDE and HCB in human milk samples with a single sample preparation. Even if the identification of these contaminants in milk has been confirmed, their exact origin is still uncertain. Further investigations are required to discriminate external contamination (e.g. occurring during sample collection, from breast pump and containers in the case of human milk) from marker of internal exposure. For the present Ph.D. project, this fine source exposure identification challenge was not addressed as it is part of the background and the priority was to develop a non-targeted workflow for characterising human samples. Finally, the data processing is still the bottleneck of the approach and recent improvements in HaloSeeker v2, iterative DDA and other software as Galaxy and MS-DIAL would help users face this challenge. Nevertheless, even if bioinformatics tools are developed, the scientific expertise is still essential. Identification would

- 208/296 - Chapter 4. Characterisation of internal human exposure to halogenated contaminants in human samples by non- targeted screening be fast and more efficient with more bioinformatics supports, however quality should remain a priority over quantity.

4.5. Analysis of meconium samples

4.5.1. Samples and protocol

In the frame of the CELPACS cohort (study conducted in the City of Brno, Czech Republic), 59 meconium samples and diapers were collected between 7th January and 28th February 2020 during a secondment included within the present Ph.D. work at RECETOX (Masaryk University, Brno, Czech Republic). For each individual sample, the entire volume of available meconium was collected in a glass tube of appropriate size. On the other hand, a piece of clean diaper was cut and stored in aluminium foil. It was proved that the cotton wool used in diaper can contains residues of pesticides (ANSES, 2019) and it could contaminate the meconium, inducing false positive results. Thus, three empty glass tubes of each size were also stored in the same conditions as samples, constituting blank containers.

A pool of the different collected meconium samples was also prepared and used as a QC sample, then expected to reflect the heterogeneity (intra- and inter-samples) of this particular matrix. After pooling all samples in the same tube, a first tentative consisted in homogenate with ultrasonic probe. This technical option was found efficient in terms of homogenisation, however it significantly altered the physical structure of meconium and led to the release of a strong odour which is abnormal because most of meconium does not smell anything. Also, the pool sample was warm after the ultrasonic disruption. This approach was then finally not retained, and a vigorous and manual homogenisation was preferred.

For this experiment, six meconium samples were prepared with the Captiva EMR-Lipids® sample preparation described earlier (Appendix 7). Samples were analysed with LC- and GC- HRMS methods detailed in Chapter 2 (sections 2.2.2 LC-HRMS method development and 2.2.3 GC-HRMS method development) and data were processed with the identification strategy detailed in the same chapter (section 2.4.2 Strategy developed to process data generated by non- targeted approach).

- 209/296 - Chapter 4. Characterisation of internal human exposure to halogenated contaminants in human samples by non- targeted screening

4.5.2. Meconium sample investigation

Investigation of LC-HRMS data

The polar fractions (acetonitrile/water fraction) resulting from our developed sample preparation applied to meconium and procedural blank samples were analysed by LC-ESI(-)- and ESI(+)-HRMS and an example of chromatograms obtained for those extracts is provided in Figure 4.10.

Figure 4.10: Overlap of total ion currents obtained for meconium (blue) and procedural blank (black) samples purified with the Captiva EMR-Lipid® protocol and analysed by LC-ESI(-)- and ESI(+)-HRMS, top and bottom respectively.

- 210/296 - Chapter 4. Characterisation of internal human exposure to halogenated contaminants in human samples by non- targeted screening

In total, 443 and 391 clusters were aligned in the six meconium samples analysed in LC-ESI(-)- and ESI(+)-HRMS, respectively, from which 121 and 80 clusters were highlighted after both blank subtraction and manual investigation. Then, based on those 41 and 14 clusters matched with the HBM4EU annotation reference database, respectively. After the investigation of these matches, none was identified as an already known contaminant. A deeper work would be required to annotate those signals with higher level of confidence, and to extend the annotation to other unknown markers of exposure. This includes the generation of MS/MS data thanks to dedicated approaches such as iDDA, as was conducted for human milk sample. However, this task is still very laborious mainly because of the lack of well-consolidated reference libraries, explaining that these LC-HRMS data generated for these six meconium samples were not further investigated. However, meconium samples constitute an interesting perspective for the present Ph.D. work since blank containers and diapers were collected as well as stool. In combination with the present non-targeted strategy and some data processing updates, those samples could be comprehensively characterised.

Investigation of GC-HRMS data

The nonpolar fraction (hexane) fraction of the same meconium extract and procedural blank samples were analysed by GC-EI-HRMS and an example of chromatograms obtained for those extracts is provided Figure 4.11.

Figure 4.11: Overlap of total ion currents obtained for meconium (blue) and procedural blank (black) samples purified with the Captiva EMR-Lipid® protocol and analysed by GC-EI- HRMS.

- 211/296 - Chapter 4. Characterisation of internal human exposure to halogenated contaminants in human samples by non- targeted screening

The generated GC-HRMS data were processed in a same way than the GC-HRMS data previously generated from human milk. The same compounds already identified in human milk samples, namely p,p’-DDE and HCB were detected in six and three meconium samples respectively. In particular, p,p’-DDE was identified with a confidence level of 1 thanks to the comparison of the fragmentation spectrum with a pure analytical standard (Figure 4.12).

Figure 4.12: Fragmentation mass spectra of p,p’-DDE in meconium samples (top) and pure analytical standard (bottom) analysed by GC-EI-HRMS.

Concentrations were estimated at maximum between 2.5 and 15 pg g-1 fresh weight. This semi- quantitative approach was based on the comparison of peak area with pure analytical standard. For the second compound, HCB, the intensity was too low to unambiguously confirm its identity. Diagnostic fragments were not detected, so it was assumed that the compound was HCB, given that it was eluted at the same retention time with a good mass accuracy (between - 1 and -0.31 ppm) for the parent ion. For the same reasons as for the investigation of GC-HRMS data on human milk, additional bioinformatics tools are required for a deeper interpretation of non-annotated signals.

- 212/296 - Chapter 4. Characterisation of internal human exposure to halogenated contaminants in human samples by non- targeted screening

4.5.3. Conclusion

The characterisation of six meconium samples led to the identification of p,p’-DDE and potential identification of HCB by GC-HRMS analysis. On the contrary, LC-HRMS data did not lead us to identify already known contaminants. This demonstrates the need for a deeper investigation of the data, in order to putatively identify non-annotated compounds, including unknown compounds.

In the same dynamic as the one previously detailed for human milk, external contamination can potentially occur during sample collection, originating from the diaper and the container used to store samples. This phenomenon was taken into account during meconium sample collections in Czech Republic. Considering the importance of this aspect, it should be commonly included in any sample collection procedure involving non-targeted characterisation, in particular for applications dedicated to the internal exposure of human to organic contaminants.

4.6. Conclusion

Real samples of human adipose tissue, placenta, human milk and meconium were characterised with the methodology developed during the present Ph.D. work including the sample preparation, analysis and data processing. The investigation of these different matrices globally allowed to confirm the efficiency of this non-targeted strategy up to the identification of several markers of human internal chemical exposure.

This application work pointed out the necessary caution with regard to the interpretation of certain signals present in procedural blank samples, thus appearing as one critical point of these open non-targeted approaches and requiring the implementation of well-defined management provisions. The prioritisation of data to be further investigated for confirmatory purpose appeared as a second critical issue associated with NTS, for which well-defined criteria should be established, for instance based on higher detection frequency. Furthermore, the data processing component of these NTS approaches appears altogether as a bottleneck, especially because of the lack of appropriate and extended annotation database, consolidated by QA/QC. In addition, new software or updates are required to process data with more accuracy.

Beyond these necessary improvements, the systematic generation and investigation of MS/MS data in LC (iterative DDA) is an interesting perspective to facilitate the annotation. In GC, the

- 213/296 - Chapter 4. Characterisation of internal human exposure to halogenated contaminants in human samples by non- targeted screening compatibility between HRMS data and the NIST in LRMS is also a promising approach for the identification.

With suitable developments and guidelines, NTS strategies will gradually enable extending the knowledge of the human exposure and characterising the pressure exerted by environmental chemicals in a more comprehensive way, according to a well-described continuum that combines environment-food-human health.

- 214/296 -

GENERAL CONCLUSION AND PERSPECTIVES

- 215/296 -

General conclusion and perspectives

The development of non-targeted screening approaches appears challenging at several points of the analytical workflow. These challenges have been addressed during the present Ph.D. project in order to propose first proofs-of-concept with regard to the application of these approaches to characterise human internal chemical exposure. It then allows to start establishing guidelines for research groups wishing to join this dynamic. The present methodology has been developed in the frame of the HBM4EU project, for a long-term contribution to the development of these new methodological strategies for human biomonitoring, environmental health studies, and support to risk assessment in Europe.

The different steps of the underlying non-targeted analytical workflow were investigated, from sample preparation to analysis and data processing. Various human matrices, including adipose tissue, placenta, human milk and meconium, were selected in order to characterise more specifically the internal human exposure to chemicals at early stage of life (i.e. during the perinatal period). The main challenge was then to capture a large range of markers of exposure featuring various physicochemical properties with sufficient sensitivity and selectivity. The sufficient removal of matrix interferences (e.g. proteins, lipids), as well as the use of low sample volumes commonly available for human matrices (i.e. 100 mg up to 1 g) accounted for the main constraining factors took into account during this development. Thus, different sample preparation protocols were assessed, namely acid hydrolysis, GPC fractionation, Bligh and Dyer partitioning and solid phase extraction on Captiva EMR-Lipid®. All the tested sample preparation options presented respective advantages and limitations which were identified and discussed. Finally, sample fractionation approaches, such as GPC, appeared particularly promising, as they comply with the global philosophy of NTS considering their comprehensiveness and preserving character with regard to the analysed sample. Protocols based on endogenous compounds removal, such as Captiva EMR-Lipid®, also appeared as a relevant option for NTS. Both approaches appear interesting and should be further investigated and may be coupled to other more selective sample preparations. Indeed, the matrix effect phenomenon appears as a main issue of concern for NTS, making this sample preparation step even more determining than for any targeted method development.

In addition, the developed sample preparation protocols should be adapted according to the used instrumentation. For the present study, a liquid-liquid partitioning was performed and resulting in polar and nonpolar fractions separately analysed by LC-ESI(+/-)- and GC-EI-HRMS, respectively. These complementary instrumental methods were developed in

- 217/296 - General conclusion and perspectives order to expand the number of accessible markers of exposure. Thus, generic settings have been optimised and this comprehensive analysis has enabled to extend the mass (100-1000 Da) and log P (from 2 to 8) ranges of detectable compounds. In addition, HRMS detectors are required to reach high mass accuracy, which is needed for compound identification. The Orbitrap technology was used in the present case, both in full-scan mode for first step screening and also in MS/MS mode (DDA) for identification purpose. For the present study, the chromatographic separation was ensured in one dimension, but as a future perspective, additional dimensions such as LCxLC, GCxGC, and/or ion mobility spectrometry also represent complementary technologies of interest. At the present time, the main limitation regarding these advanced analytical approaches may be the lack of really adapted software able to process such complex data, which hampers their widespread application.

Last but not least, data processing is probably the most challenging step of the NTS workflows as it does not only require analytical knowledge and expertise, but also advanced bioinformatics skills not historically nor yet well implemented in laboratories. Data processing is still facing difficulties to highlight signals of potential interest (i.e. useful information) from the vast volume of crude data generated by these approaches. The developed software HaloSeeker was used to more specifically focus on chlorinated and brominated substances, as one of possible prioritisation key. During the present Ph.D., additional tools to identify and manage signals resulting from external contamination were also implemented (e.g. alignment, blank subtraction). It allowed with the dedicated developed strategy to focus the investigation on potential exogenous compounds, step-by-step. Furthermore, molecular networks, already implemented in metabolomics workflows, appear as another promising tool that could be used to filter signal of potential interest, for instance, by grouping signals related to matrix interferences (e.g. lipids) or functionally linked markers. Finally, once a signal was considered of interest, the annotation of the captured markers of exposure was carried out using the HBM4EU annotation reference database. However, this database is still under construction and in the current state is still lacking information (such as MS/MS spectra) to unambiguously confirm the identification of all markers. Importantly, other approaches as iterative DDA have to be further developed and consolidated to efficiently generate structural information on unannotated signals and contribute to this identification step. Finally, the unambiguous identification will be possible if the analytical standard corresponding to the considered marker is available. We can easily imagine that is not the case for unknown compounds, for which additional research would be needed to elucidate their structure.

- 218/296 - General conclusion and perspectives

Regarding the application of the developed technical capabilities for NTS, some markers of chemical exposure were successfully identified thanks to the expertise gained during the present project and pluridisciplinary collaborations. α-HBCDD, triclosan and 4-hydroxy-chlorothalonil were identified by LC-HRMS in adipose tissue, placenta and human milk, respectively, alongside p,p’-DDE and HCB were identified by GC-HRMS in human milk and meconium. These first proofs-of-concept demonstrate the strengths of NTS approaches to seek organic contaminants in human matrices without a priori knowledge. They also illustrate the challenges faced to generate such results, as the few identified compounds probably represent a small part of all contaminants present in the sample. The next step is to investigate more deeply the already acquired data in order to pursue the task of data mining and identification of markers. Furthermore, it is important to continue to generate new data of exposure on real samples by these approaches as the accumulation of such data may also contribute to progress on the prioritisation and identification of markers of interest. Once these NTS data are acquired, they can be stored for years and reprocessed with new tools, which also represent a strength of these approaches.

In a frame of the collaboration with Vrije Universiteit, another approach to screen for potential markers of exposure of interest have been assessed, based on biological activity testing. This EDA approach allowed us to combine the chemical approach developed above with a biological test focused on thyroid hormone system disruptors. Biological approaches also appeared of particular interest for charactering mixture effect. However, the sensitivity of this approach is not yet compatible with expected contamination levels in human samples. Hence, further developments are needed. This complementary approach to the NTS approach developed during the present Ph.D. work illustrates the need of collaboration between laboratories. Indeed, EDA requires specific equipment, knowledge and skills in biology and toxicology, which is not directly implementable in all analytical laboratories.

The assessment of the non-targeted method performance was another important component of the present Ph.D. work. The matrix effect phenomenon was studied with qualitative and quantitative approaches and we advise to systematically include this experiment in the process of development of non-targeted methods. A dual characterisation of the same samples with both targeted and non-targeted analyses also appeared interesting during this first developing age of NTS, in order to i) better characterise the existing gap between both approaches in terms of specificity and sensitivity and ii) to determine if not detected compounds are either not present

- 219/296 - General conclusion and perspectives in the sample or not recovered by the non-targeted methods. Then, the present study confirmed the low contamination level of human samples with regard to some markers of chemical exposure, and the still limiting difference of sensitivity between both targeted and non-targeted approaches. It was shown that the non-targeted approach can be presently used to characterise samples with the highest concentration levels i.e. the most highly exposed individuals. On one hand, our overall results confirmed the relevance and compatibility of non-targeted methods and their application for human biomonitoring. On the other hand, it demonstrated that further advances are still required to improve their sensitivity and high throughput capacities.

The management of the external contamination possibly occurring during the sample collection, preparation and analysis is another crucial aspect of the non-targeted approaches that still raises interest, and for which significant harmonisation is required at international level. For instance, blank containers could be systemically included during sample collection, as it was done during the present Ph.D. work in collaboration with RECETOX to collect meconium samples. As a minimal requirement, field and procedural blank samples have to be prepared and analysed in parallel of matrix samples. Then, those blank samples should be investigated in more detail to assess the repeatability of the external contamination and then be able to detect ubiquitous contaminants.

To conclude, our results have demonstrated that the analytical strategy developed is effective for the non-targeted screening of markers of human internal chemical exposure. This emerging approach is currently a matter of extremely high interest, wide expectations, and intense technical developments at international level. At a research level, they are promised to very quickly produce a new generation of extended internal exposure data and contribute to the widening of knowledge of the human exposome and to offer new perspectives for environmental health studies. At a regulatory level, in the context of human biomonitoring and on the long term, non-targeted approaches may represent a new tool for early warning support to policy and prioritisation of further HBM programs and risk assessment. However, this emerging field requires an important effort in terms of harmonisation at international level for anticipating further data comparability and to consolidate result interpretation. It should be complementary to other initiatives in the environmental and food safety areas where these approaches are also developed. This implies to export non-targeted methods in all continents, including countries with reduced access to last generation of instrumentations and software

- 220/296 - General conclusion and perspectives which compromise the implementation of the strategy. This implies to initiate international collaborations, akin to the HBM4EU project and beyond Europe.

- 221/296 -

LIST OF BIBLIOGRAPHICAL REFERENCES

- 223/296 -

List of bibliographical references

Abate-Pella, D.; Freund, D. M.; Ma, Y.; Simón-Manso, Y.; Hollender, J.; Broeckling, C. D.; Huhman, D. V.; Krokhin, O. V.; Stoll, D. R.; Hegeman, A. D.; Kind, T.; Fiehn, O.; Schymanski, E. L.; Prenni, J. E.; Sumner, L. W.; Boswell, P. G., Retention projection enables accurate calculation of liquid chromatographic retention times across labs and methods. Journal of Chromatography A 2015, 1412, 43-51.

Aerts, R.; Van Overmeire, I.; Colles, A.; Andjelković, M.; Malarvannan, G.; Poma, G.; Den Hond, E.; Van de Mieroop, E.; Dewolf, M.-C.; Charlet, F.; Van Nieuwenhuyse, A.; Van Loco, J.; Covaci, A., Determinants of persistent organic pollutant (POP) concentrations in human breast milk of a cross-sectional sample of primiparous mothers in Belgium. Environment International 2019, 131, 104979.

Alexandre-Gouabau, M. C.; Moyon, T.; David-Sochard, A.; Fenaille, F.; Cholet, S.; Royer, A. L.; Guitton, Y.; Billard, H.; Darmaun, D.; Roze, J. C.; Boquien, C. Y., Comprehensive Preterm Breast Milk Metabotype Associated with Optimal Infant Early Growth Pattern. Nutrients 2019, 11 (3).

Alves, A.; Kucharska, A.; Erratico, C.; Xu, F.; Den Hond, E.; Koppen, G.; Vanermen, G.; Covaci, A.; Voorspoels, S., Human biomonitoring of emerging pollutants through non- invasive matrices: state of the art and future potential. Analytical and Bioanalytical Chemistry 2014, 406 (17), 4063-4088.

Angerer, J.; Ewers, U.; Wilhelm, M., Human biomonitoring: State of the art. International Journal of Hygiene and Environmental Health 2007, 210 (3), 201-228.

ANSES, Safety of baby diapers – ANSES revised opinion Collective expert appraisal report, January 2019

Antignac, J.-P.; de Wasch, K.; Monteau, F.; De Brabander, H.; Andre, F.; Le Bizec, B., The ion suppression phenomenon in liquid chromatography–mass spectrometry and its consequences in the field of residue analysis. Analytica Chimica Acta 2005, 529 (1-2), 129- 136.

Arce-López, B.; Lizarraga, E.; Flores-Flores, M.; Irigoyen, Á.; González-Peñas, E., Development and validation of a methodology based on Captiva EMR-lipid clean-up and LC-

- 225/296 - List of bibliographical references

MS/MS analysis for the simultaneous determination of mycotoxins in human plasma. Talanta 2020, 206, 120193.

Bakhytkyzy, I.; Hewelt-Belka, W.; Kot-Wasik, A., The dispersive micro-solid phase extraction method for MS-based lipidomics of human breast milk. Microchemical Journal 2020, 152, 104269.

Baran, R., Untargeted metabolomics suffers from incomplete raw data processing. Metabolomics 2017, 13 (9), 107.

Barr, D. B.; Wang, R. Y.; Needham, L. L., Biologic Monitoring of Exposure to Environmental Chemicals throughout the Life Stages: Requirements and Issues for Consideration for the National Children’s Study. Environmental Health Perspectives 2005, 113 (8), 1083-1091.

Bates, M. N.; Hamilton, J. W.; LaKind, J. S.; Langenberg, P.; O’Malley, M.; Snodgrass, W., Workgroup Report: Biomonitoring Study Design, Interpretation, and Communication - Lessons Learned and Path Forward. Environmental Health Perspectives 2005, 113 (11), 1615- 1621.

Berghuis, S. A.; Van Braeckel, K. N. J. A.; Sauer, P. J. J.; Bos, A. F., Prenatal exposure to persistent organic pollutants and cognition and motor performance in adolescence. Environment International 2018, 121, 13-22.

Bichon, E.; Guiffard, I.; Vénisseau, A.; Lesquin, E.; Vaccher, V.; Marchand, P.; Le Bizec, B., Simultaneous analysis of historical, emerging and novel brominated flame retardants in food and feed using a common extraction and purification method. Chemosphere 2018, 205, 31-40.

Bichon, E.; Guiffard, I.; Vénisseau, A.; Marchand, P.; Antignac, J.-P.; Le Bizec, B., Ultra-trace quantification method for chlordecone in human fluids and tissues. Journal of Chromatography A 2015, 1408 (Supplement C), 169-177.

Bligh, E. G.; Dyer, W. J., A rapid method of total lipid extraction and purification. Canadian journal of biochemistry and physiology 1959, 37 (8), 911-917.

- 226/296 - List of bibliographical references

Boccard, J.; Veuthey, J. L.; Rudaz, S., Knowledge discovery in metabolomics: an overview of MS data handling. Journal of separation science 2010, 33 (3), 290-304.

Booij, P.; Vethaak, A. D.; Leonards, P. E. G.; Sjollema, S. B.; Kool, J.; de Voogt, P.; Lamoree, M. H., Identification of Photosynthesis Inhibitors of Pelagic Marine Algae Using 96- Well Plate Microfractionation for Enhanced Throughput in Effect-Directed Analysis. Environmental Science & Technology 2014, 48 (14), 8003-8011.

Brack, W.; Ait-Aissa, S.; Burgess, R. M.; Busch, W.; Creusot, N.; Di Paolo, C.; Escher, B. I.; Mark Hewitt, L.; Hilscherova, K.; Hollender, J.; Hollert, H.; Jonker, W.; Kool, J.; Lamoree, M.; Muschket, M.; Neumann, S.; Rostkowski, P.; Ruttkies, C.; Schollee, J.; Schymanski, E. L.; Schulze, T.; Seiler, T.-B.; Tindall, A. J.; De Aragão Umbuzeiro, G.; Vrana, B.; Krauss, M., Effect-directed analysis supporting monitoring of aquatic environments — An in-depth overview. Science of The Total Environment 2016, 544 (Supplement C), 1073-1118.

Buekers, J.; David, M.; Koppen, G.; Bessems, J.; Scheringer, M.; Lebret, E.; Sarigiannis, D.; Kolossa-Gehring, M.; Berglund, M.; Schoeters, G.; Trier, X., Development of Policy Relevant Human Biomonitoring Indicators for Chemical Exposure in the European Population. Int J Environ Res Public Health 2018, 15 (10), 2085.

Caesar, L. K.; Kvalheim, O. M.; Cech, N. B., Hierarchical cluster analysis of technical replicates to identify interferents in untargeted mass spectrometry metabolomics. Analytica Chimica Acta 2018, 1021, 69-77.

Cajka, T.; Fiehn, O., Toward merging untargeted and targeted methods in mass spectrometry-based metabolomics and lipidomics. Analytical chemistry 2016, 88 (1), 524-545.

Campoy, C.; Jiménez, M.; Olea-Serrano, M. F.; Moreno Frias, M.; Cañabate, F.; Olea, N.; Bayés, R.; Molina-Font, J. A., Analysis of organochlorine pesticides in human milk: preliminary results. Early Human Development 2001, 65, S183-S190.

Cano-Sancho, G.; Alexandre-Gouabau, M.-C.; Moyon, T.; Royer, A.-L.; Guitton, Y.; Billard, H.; Darmaun, D.; Rozé, J.-C.; Boquien, C.-Y.; Le Bizec, B.; Antignac, J.-P., Simultaneous exploration of nutrients and pollutants in human milk and their impact on preterm infant growth: An integrative cross-platform approach. Environmental Research 2020, 182, 109018.

- 227/296 - List of bibliographical references

Cariou, R.; Omer, E.; Léon, A.; Dervilly-Pinel, G.; Le Bizec, B., Screening halogenated environmental contaminants in biota based on isotopic pattern and mass defect provided by high resolution mass spectrometry profiling. Analytica Chimica Acta 2016, 936 (Supplement C), 130-138.

Carson, R., Silent spring - 40th Anniversary edition. Mariner; 2002

Carvalho, F. P., Pesticides, environment, and food safety. Food and Energy Security 2017, 6 (2), 48-60.

Cassoulet, R.; Haroune, L.; Abdelouahab, N.; Gillet, V.; Baccarelli, A. A.; Cabana, H.; Takser, L.; Bellenger, J.-P., Monitoring of prenatal exposure to organic and inorganic contaminants using meconium from an Eastern Canada cohort. Environmental Research 2019, 171, 44-51.

Celma, A.; Bijlsma, L.; López, F. J.; Sancho, J. V., Development of a Retention Time Interpolation scale (RTi) for liquid chromatography coupled to mass spectrometry in both positive and negative ionization modes. Journal of Chromatography A 2018, 1568, 101-107.

Cloutier, P.-L.; Fortin, F.; Groleau, P. E.; Brousseau, P.; Fournier, M.; Desrosiers, M., QuEChERS extraction for multi-residue analysis of PCBs, PAHs, PBDEs and PCDD/Fs in biological samples. Talanta 2017, 165, 332-338.

Considine, E.; Thomas, G.; Boulesteix, A.-L.; Khashan, A.; Kenny, L., Critical review of reporting of the data analysis step in metabolomics. Metabolomics 2018, 14 (1), 7.

Cortéjade, A. l.; Kiss, A.; Cren, C.; Vulliet, E.; Buleté, A., Development of an analytical method for the targeted screening and multi-residue quantification of environmental contaminants in urine by liquid chromatography coupled to high resolution mass spectrometry for evaluation of human exposures. Talanta 2016, 146, 694-706.

Côté, C.; Bergeron, A.; Mess, J.-N.; Furtado, M.; Garofolo, F., Matrix effect elimination during LC–MS/MS bioanalytical method development. Bioanalysis 2009, 1 (7), 1243-1257.

Couchman, L., Turbulent flow chromatography in bioanalysis: a review. Biomedical Chromatography 2012, 26 (8), 892-905.

- 228/296 - List of bibliographical references

Dadvand, P.; Parker, J.; Bell, M. L.; Bonzini, M.; Brauer, M.; Darrow, L. A.; Gehring, U.; Glinianaia, S. V.; Gouveia, N.; Ha, E.-h.; Leem, J. H.; Hooven, E. H. v. d.; Jalaludin, B.; Jesdale, B. M.; Lepeule, J.; Morello-Frosch, R.; Morgan, G. G.; Pesatori, A. C.; Pierik, F. H.; Pless-Mulloli, T.; Rich, D. Q.; Sathyanarayana, S.; Seo, J.; Slama, R.; Strickland, M.; Tamburic, L.; Wartenberg, D.; Nieuwenhuijsen, M. J.; Woodruff, T. J., Maternal Exposure to Particulate Air Pollution and Term Birth Weight: A Multi-Country Evaluation of Effect and Heterogeneity. Environmental Health Perspectives 2013, 121 (3), 267-373

Dennis, K. K.; Marder, E.; Balshaw, D. M.; Cui, Y.; Lynes, M. A.; Patti, G. J.; Rappaport, S. M.; Shaughnessy, D. T.; Vrijheid, M.; Barr, D. B., Biomonitoring in the era of the exposome. Environmental health perspectives 2016, 125 (4), 502-510.

Drakvik, E.; Altenburger, R.; Aoki, Y.; Backhaus, T.; Bahadori, T.; Barouki, R.; Brack, W.; Cronin, M. T. D.; Demeneix, B.; Hougaard Bennekou, S.; van Klaveren, J.; Kneuer, C.; Kolossa-Gehring, M.; Lebret, E.; Posthuma, L.; Reiber, L.; Rider, C.; Rüegg, J.; Testa, G.; van der Burg, B.; van der Voet, H.; Warhurst, A. M.; van de Water, B.; Yamazaki, K.; Öberg, M.; Bergman, Å., Statement on advancing the assessment of chemical mixtures and their risks for human health and the environment. Environment International 2020, 134, 105267.

Dudzik, D.; Barbas-Bernardos, C.; García, A.; Barbas, C., Quality assurance procedures for mass spectrometry untargeted metabolomics. A review. Journal of pharmaceutical and biomedical analysis 2018.

ECHA Information on chemicals. https://echa.europa.eu/information-on- chemicals/pic/chemicals.

Elert, M.; Bonnard, R.; Jones, C.; Schoof, R. A.; Swartjes, F. A., Human Exposure Pathways. In Dealing with Contaminated Sites: From Theory towards Practical Application, Swartjes, F. A., Ed. Springer Netherlands: Dordrecht, 2011; pp 455-515.

Eriksson, L.; Trygg, J.; Wold, S., CV-ANOVA for significance testing of PLS and OPLS® models. Journal of Chemometrics 2008, 22 (11‐12), 594-600.

Esteban, M.; Castaño, A., Non-invasive matrices in human biomonitoring: A review. Environment International 2009, 35 (2), 438-449.

- 229/296 - List of bibliographical references

EURL, Modified QuEChERS-Method for the Analysis of Chlorothalonil in Fruits and Vegetables; EU Reference Laboratories for Residues of Pesticides: 2010

Fernández-Peralbo, M.; Luque de Castro, M., Preparation of urine samples prior to targeted or untargeted metabolomics mass-spectrometry analysis. TrAC Trends in Analytical Chemistry 2012, 41, 75-85.

Folch, J.; Lees, M.; Stanley, G. S., A simple method for the isolation and purification of total lipides from animal tissues. Journal of biological chemistry 1957, 226 (1), 497-509.

Frederiksen, H.; Skakkebaek, N. E.; Andersson, A.-M., Metabolism of phthalates in humans. Molecular Nutrition & Food Research 2007, 51 (7), 899-911.

Fry, K.; Power, M. C., Persistent organic pollutants and mortality in the United States, NHANES 1999–2011. Environmental Health 2017, 16 (1), 105.

Fürst, P., Dioxins, polychlorinated biphenyls and other organohalogen compounds in human milk. Levels, correlations, trends and exposure through breastfeeding. Molecular Nutrition & Food Research 2006, 50 (10), 922-933.

Ganzleben, C.; Antignac, J.-P.; Barouki, R.; Castaño, A.; Fiddicke, U.; Klánová, J.; Lebret, E.; Olea, N.; Sarigiannis, D.; Schoeters, G. R., Human biomonitoring as a tool to support chemicals regulation in the European Union. International journal of hygiene and environmental health 2017, 220 (2 Pt A), 94.

Garwolińska, D.; Namieśnik, J.; Kot-Wasik, A.; Hewelt-Belka, W., State of the art in sample preparation for human breast milk metabolomics—merits and limitations. TrAC Trends in Analytical Chemistry 2019, 114, 1-10.

Giacomoni, F.; Le Corguillé, G.; Monsoor, M.; Landi, M.; Pericard, P.; Pétéra, M.; Duperier, C.; Tremblay-Franco, M.; Martin, J.-F.; Jacob, D.; Goulitquer, S.; Thévenot, E. A.; Caron, C., Workflow4Metabolomics: a collaborative research infrastructure for computational metabolomics. Bioinformatics 2014, 31 (9), 1493-1495.

Giardine, B.; Riemer, C.; Hardison, R. C.; Burhans, R.; Elnitski, L.; Shah, P.; Zhang, Y.; Blankenberg, D.; Albert, I.; Taylor, J.; Miller, W.; Kent, W. J.; Nekrutenko, A., Galaxy: A

- 230/296 - List of bibliographical references platform for interactive large-scale genome analysis. Genome Research 2005, 15 (10), 1451- 1455.

Göen, T.; Schaller, K.-H.; Drexler, H., External quality assessment of human biomonitoring in the range of environmental exposure levels. International Journal of Hygiene and Environmental Health 2012, 215 (2), 229-232.

Guillarme, D.; Veuthey, J.-L., Chapter 1 - Theory and Practice of UHPLC and UHPLC–MS. In Handbook of Advanced Chromatography/Mass Spectrometry Techniques, Holčapek, M.; Byrdwell, W. C., Eds. AOCS Press: 2017; pp 1-38.

Gützkow, K. B.; Haug, L. S.; Thomsen, C.; Sabaredzovic, A.; Becher, G.; Brunborg, G., Placental transfer of perfluorinated compounds is selective – A Norwegian Mother and Child sub-cohort study. International Journal of Hygiene and Environmental Health 2012, 215 (2), 216-219.

Guxens, M.; Ballester, F.; Espada, M.; Fernández, M. F.; Grimalt, J. O.; Ibarluzea, J.; Olea, N.; Rebagliato, M.; Tardón, A.; Torrent, M.; Vioque, J.; Vrijheid, M.; Sunyer, J.; Project, o. b. o. t. I., Cohort Profile: The INMA—INfancia y Medio Ambiente—(Environment and Childhood) Project. International Journal of Epidemiology 2011, 41 (4), 930-940.

Han, L.; Matarrita, J.; Sapozhnikova, Y.; Lehotay, S. J., Evaluation of a recent product to remove lipids and other matrix co-extractives in the analysis of pesticide residues and environmental contaminants in foods. Journal of Chromatography A 2016, 1449, 17-29.

HBM4EU, Deliverable report AD14.4, First report on the state of development of new biomarkers of effect – WP 14 Biomarkers of effect, 10 January 2019

Herrera‐Lopez, S.; Hernando, M.; García‐Calvo, E.; Fernández‐Alba, A.; Ulaszewska, M., Simultaneous screening of targeted and non‐targeted contaminants using an LC‐QTOF‐MS system and automated MS/MS library searching. Journal of Mass Spectrometry 2014, 49 (9), 878-893.

Hoffmann, E. d.; Stroobant, V., Mass spectrometry: Principles and Applications. Third edition. John Wiley & Sons: 2007.

- 231/296 - List of bibliographical references

Hu, M.; Krauss, M.; Brack, W.; Schulze, T., Optimization of LC-Orbitrap-HRMS acquisition and MZmine 2 data processing for nontarget screening of environmental samples using design of experiments. Analytical and bioanalytical chemistry 2016, 408 (28), 7905-7915.

Hübschmann, H.-J., Handbook of GC-MS: Fundamentals and Applications. Third edition. Wiley-VCH: 2015.

Inthavong, C.; Hommet, F.; Bordet, F.; Rigourd, V.; Guérin, T.; Dragacci, S., Simultaneous liquid chromatography–tandem mass spectrometry analysis of brominated flame retardants (tetrabromobisphenol A and hexabromocyclododecane diastereoisomers) in French breast milk. Chemosphere 2017, 186, 762-769.

Iribarne-Durán, L. M.; Domingo-Piñar, S.; Peinado, F. M.; Vela-Soria, F.; Jiménez- Díaz, I.; Barranco, E.; Olea, N.; Freire, C.; Artacho-Cordón, F.; Ocón-Hernández, O., Menstrual blood concentrations of parabens and benzophenones and related factors in a sample of Spanish women: An exploratory study. Environmental Research 2020, 183, 109228.

Jandera, P.; Janás, P., Recent advances in stationary phases and understanding of retention in hydrophilic interaction chromatography. A review. Analytica chimica acta 2017, 967, 12-32.

Jones, D. P., Sequencing the exposome: a call to action. Toxicology reports 2016, 3, 29-45.

Junot, C.; Fenaille, F.; Colsch, B.; Bécher, F., High resolution mass spectrometry based techniques at the crossroads of metabolic pathways. Mass Spectrometry Reviews 2014, 33 (6), 471-500.

Kendrick, E., A Mass Scale Based on CH2 = 14.0000 for High Resolution Mass Spectrometry of Organic Compounds. Analytical Chemistry 1963, 35 (13), 2146-2154.

Khamis, M. M.; Adamko, D. J.; El‐Aneed, A., Mass spectrometric based approaches in urine metabolomics and biomarker discovery. Mass spectrometry reviews 2017, 36 (2), 115- 134.

- 232/296 - List of bibliographical references

Kind, T.; Fiehn, O., Metabolomic database annotations via query of elemental compositions: Mass accuracy is insufficient even at less than 1 ppm. BMC Bioinformatics 2006, 7 (1), 234.

Koelmel, J. P.; Kroeger, N. M.; Gill, E. L.; Ulmer, C. Z.; Bowden, J. A.; Patterson, R. E.; Yost, R. A.; Garrett, T. J., Expanding Lipidome Coverage Using LC-MS/MS Data- Dependent Acquisition with Automated Exclusion List Generation. J Am Soc Mass Spectrom 2017, 28 (5), 908-917.

Kolossa-Gehring, M.; Fiddicke, U.; Leng, G.; Angerer, J.; Wolz, B., New human biomonitoring methods for chemicals of concern—the German approach to enhance relevance. International Journal of Hygiene and Environmental Health 2017, 220 (2, Part A), 103-112.

König, K.; Goethel, S. F.; Rusu, V. M.; Vogeser, M., Deproteination of serum samples for LC–MS/MS analyses by applying magnetic micro-particles. Clinical Biochemistry 2013, 46 (7), 652-655.

Kováts, E., Gas-chromatographische Charakterisierung organischer Verbindungen. Teil 1: Retentionsindices aliphatischer Halogenide, Alkohole, Aldehyde und Ketone. Helvetica Chimica Acta 1958, 41 (7), 1915-1932.

Kruve, A.; Kaupmees, K., Adduct Formation in ESI/MS by Mobile Phase Additives. Journal of The American Society for Mass Spectrometry 2017, 28 (5), 887-894.

Kuhl, C.; Tautenhahn, R.; Böttcher, C.; Larson, T. R.; Neumann, S., CAMERA: An Integrated Strategy for Compound Spectra Extraction and Annotation of Liquid Chromatography/Mass Spectrometry Data Sets. Analytical Chemistry 2012, 84 (1), 283-289.

Kwiecien, N. W.; Bailey, D. J.; Rush, M. J. P.; Cole, J. S.; Ulbrich, A.; Hebert, A. S.; Westphall, M. S.; Coon, J. J., High-Resolution Filtering for Improved Small Molecule Identification via GC/MS. Analytical Chemistry 2015, 87 (16), 8328-8335

Kyriklaki, A.; Vafeiadi, M.; Kampouri, M.; Koutra, K.; Roumeliotaki, T.; Chalkiadaki, G.; Anousaki, D.; Rantakokko, P.; Kiviranta, H.; Fthenou, E.; Bitsios, P.; Kyrtopoulos, S. A.; Kogevinas, M.; Chatzi, L., Prenatal exposure to persistent organic pollutants in association with offspring neuropsychological development at 4years of age: The Rhea mother-child cohort, Crete, Greece. Environment International 2016, 97, 204-211.

- 233/296 - List of bibliographical references

La Merrill, M.; Emond, C.; Kim, M. J.; Antignac, J.-P.; Le Bizec, B.; Clément, K.; Birnbaum, L. S.; Barouki, R., Toxicological Function of Adipose Tissue: Focus on Persistent Organic Pollutants. Environmental Health Perspectives 2013, 121 (2), 162-169.

Lai, Z.; Tsugawa, H.; Wohlgemuth, G.; Mehta, S.; Mueller, M.; Zheng, Y.; Ogiwara, A.; Meissen, J.; Showalter, M.; Takeuchi, K.; Kind, T.; Beal, P.; Arita, M.; Fiehn, O., Identifying metabolites by integrating metabolome databases with mass spectrometry cheminformatics. Nature Methods 2018, 15 (1), 53-56.

Lee, Y.-M.; Kim, K.-S.; Jacobs Jr, D. R.; Lee, D.-H., Persistent organic pollutants in adipose tissue should be considered in obesity research. Obesity Reviews 2017, 18 (2), 129- 139.

Leino, O.; Kiviranta, H.; Karjalainen, A. K.; Kronberg-Kippilä, C.; Sinkko, H.; Larsen, E. H.; Virtanen, S.; Tuomisto, J. T., Pollutant concentrations in placenta. Food and Chemical Toxicology 2013, 54, 59-69.

Léon, A.; Cariou, R.; Hutinet, S.; Hurel, J.; Guitton, Y.; Tixier, C. l.; Munschy, C.; Antignac, J.-P.; Dervilly-Pinel, G.; Le Bizec, B., HaloSeeker 1.0: A User-Friendly Software to Highlight Halogenated Chemicals in Nontargeted High-Resolution Mass Spectrometry Data Sets. Analytical chemistry 2019, 91 (5), 3500-3507.

Li Q. Q.; Loganath, A.; Chong, Y. S.; Tan, J.; Obbard, J. P., Persistent Organic Pollutants and Adverse Health Effects in Humans. Journal of Toxicology and Environmental Health, Part A 2006, 69 (21), 1987-2005.

Lin, L.; Yu, Q.; Yan, X.; Hang, W.; Zheng, J.; Xing, J.; Huang, B., Direct infusion mass spectrometry or liquid chromatography mass spectrometry for human metabonomics? A serum metabonomic study of kidney cancer. Analyst 2010, 135 (11), 2970-2978.

Llompart, M.; Celeiro, M.; Dagnac, T., Microwave-assisted extraction of pharmaceuticals, personal care products and industrial contaminants in the environment. TrAC Trends in Analytical Chemistry 2019, 116, 136-150.

Lommen, A.; Kools, H. J., MetAlign 3.0: performance enhancement by efficient use of advances in computer hardware. Metabolomics 2012, 8 (4), 719-726.

- 234/296 - List of bibliographical references

López-Blanco, R.; Nortes-Mendez, R.; Robles-Molina, J.; Moreno-Gonzalez, D.; Gilbert-Lopez, B.; Garcia-Reyes, J. F.; Molina-Diaz, A., Evaluation of different cleanup sorbents for multiresidue pesticide analysis in fatty vegetable matrices by liquid chromatography tandem mass spectrometry. Journal of Chromatography A 2016, 1456, 89-104.

Losacco, G. L.; Veuthey, J.-L.; Guillarme, D., Supercritical fluid chromatography – Mass spectrometry: Recent evolution and current trends. TrAC Trends in Analytical Chemistry 2019, 118, 731-738.

Louro, H.; Heinälä, M.; Bessems, J.; Buekers, J.; Vermeire, T.; Woutersen, M.; van Engelen, J.; Borges, T.; Rousselle, C.; Ougier, E.; Alvito, P.; Martins, C.; Assuncao, R.; Joao Silva, M.; Pronk, A.; Schaddelee-Scholten, B.; Del Carmen Gonzalez, M.; De Alba, M.; Castano, A.; Viegas, S.; Humar-Juric, T.; Kononenko, L.; Lampen, A.; Vinggaard, A. M.; Schoeters, G.; Kolossa-Gehring, M.; Santonen, T., Human biomonitoring in health risk assessment in Europe: Current practices and recommendations for the future. International journal of hygiene and environmental health 2019.

Ludwig, C.; Gillet, L.; Rosenberger, G.; Amon, S.; Collins, B. C.; Aebersold, R., Data‐ independent acquisition‐based SWATH‐MS for quantitative proteomics: a tutorial. Molecular systems biology 2018, 14 (8).

Lung, D.; Wilson, N.; Chatenet, F.-T.; LaCroix, C.; Gerona, R., Non-targeted screening for novel psychoactive substances among agitated emergency department patients. Clinical Toxicology 2016, 54 (4), 319-323.

Mardal, M.; Andreasen, M. F.; Mollerup, C. B.; Stockham, P.; Telving, R.; Thomaidis, N. S.; Diamanti, K. S.; Linnet, K.; Dalsgaard, P. W., HighResNPS. com: An Online Crowd- Sourced HR-MS Database for Suspect and Non-targeted Screening of New Psychoactive Substances. Journal of analytical toxicology 2019, 1-8.

Matyash, V.; Liebisch, G.; Kurzchalia, T. V.; Shevchenko, A.; Schwudke, D., Lipid extraction by methyl-tert-butyl ether for high-throughput lipidomics. Journal of lipid research 2008, 49 (5), 1137-1146.

- 235/296 - List of bibliographical references

McCord, J.; Strynar, M., Identifying Per-and Polyfluorinated Chemical Species with a Combined Targeted and Non-Targeted-Screening High-Resolution Mass Spectrometry Workflow. JoVE (Journal of Visualized Experiments) 2019, (146), e59142.

McEachran, A. D.; Mansouri, K.; Grulke, C.; Schymanski, E. L.; Ruttkies, C.; Williams, A. J., “MS-Ready” structures for non-targeted high-resolution mass spectrometry screening studies. Journal of cheminformatics 2018, 10 (1), 45.

McEachran, A. D.; Sobus, J. R.; Williams, A. J., Identifying known unknowns using the US EPA’s CompTox Chemistry Dashboard. Analytical and bioanalytical chemistry 2017, 409 (7), 1729-1735.

Mitchell, J. M.; Fan, T. W.-M.; Lane, A. N.; Moseley, H. N. B., Development and in silico evaluation of large-scale metabolite identification methods using functional group detection for metabolomics. Frontiers in Genetics 2014, 5 (237).

Mustieles, V.; Fernández, M. F.; Martin-Olmedo, P.; González-Alzaga, B.; Fontalba- Navas, A.; Hauser, R.; Olea, N.; Arrebola, J. P., Human adipose tissue levels of persistent organic pollutants and metabolic syndrome components: Combining a cross-sectional with a 10-year longitudinal study using a multi-pollutant approach. Environment International 2017, 104, 48-57.

Myren, M.; Mose, T.; Mathiesen, L.; Knudsen, L. E., The human placenta – An alternative for studying foetal exposure. Toxicology in Vitro 2007, 21 (7), 1332-1340.

Nanes, J. A.; Xia, Y.; Dassanayake, R. M. A. P. S.; Jones, R. M.; Li, A.; Stodgell, C. J.; Walker, C. K.; Szabo, S.; Leuthner, S.; Durkin, M. S.; Moye, J.; Miller, R. K., Selected persistent organic pollutants in human placental tissue from the United States. Chemosphere 2014, 106, 20-27.

Narduzzi, L.; Royer, A.-L.; Bichon, E.; Guitton, Y.; Buisson, C.; Le Bizec, B.; Dervilly-Pinel, G., Ammonium Fluoride as Suitable Additive for HILIC-Based LC-HRMS Metabolomics. Metabolites 2019, 9 (12), 292.

Needham, L. L.; Barr, D. B.; Calafat, A. M., Characterizing children's exposures: beyond NHANES. NeuroToxicology 2005, 26 (4), 547-553.

- 236/296 - List of bibliographical references

Nickerson, K., Environmental Contaminants in Breast Milk. Journal of Midwifery & Women's Health 2006, 51 (1), 26-34.

Niedzwiecki, M. M.; Walker, D. I.; Vermeulen, R.; Chadeau-Hyam, M.; Jones, D. P.; Miller, G. W., The Exposome: Molecules to Populations. Annual Review of Pharmacology and Toxicology 2019, 59 (1), 107-127.

Oberacher, H.; Arnhard, K., Compound identification in forensic toxicological analysis with untargeted LC–MS-based techniques. Bioanalysis 2015, 7 (21), 2825-2840.

Oberacher, H.; Reinstadler, V.; Kreidl, M.; Stravs, M.; Hollender, J.; Schymanski, E., Annotating Nontargeted LC-HRMS/MS Data with Two Complementary Tandem Mass Spectral Libraries. Metabolites 2019, 9 (1), 3.

Oddy, W. H., Breastfeeding protects against illness and infection in infants and children: a review of the evidence. Breastfeeding Review 2001, 9 (2), 8-11.

Omer, E.; Cariou, R.; Remaud, G.; Guitton, Y.; Germon, H.; Hill, P.; Dervilly-Pinel, G.; Le Bizec, B., Elucidation of non-intentionally added substances migrating from polyester- polyurethane lacquers using automated LC-HRMS data processing. Analytical and Bioanalytical Chemistry 2018, 410 (22), 5391-5403.

Ortega García, J. A.; Carrizo Gallardo, D.; Ferris i Tortajada, J.; García, M. M. P.; Grimalt, J. O., Meconium and neurotoxicants: searching for a prenatal exposure timing. Archives of Disease in Childhood 2006, 91 (8), 642-646.

Ouyang, X.; Froment, J.; Leonards, P. E. G.; Christensen, G.; Tollefsen, K.-E.; de Boer, J.; Thomas, K. V.; Lamoree, M. H., Miniaturization of a transthyretin binding assay using a fluorescent probe for high throughput screening of thyroid hormone disruption in environmental samples. Chemosphere 2017, 171, 722-728.

Papadopoulou, E.; Padilla-Sanchez, J. A.; Collins, C. D.; Cousins, I. T.; Covaci, A.; de Wit, C. A.; Leonards, P. E. G.; Voorspoels, S.; Thomsen, C.; Harrad, S.; Haug, L. S., Sampling strategy for estimating human exposure pathways to consumer chemicals. Emerging Contaminants 2016, 2 (1), 26-36.

- 237/296 - List of bibliographical references

Patterson, R. E.; Ducrocq, A. J.; McDougall, D. J.; Garrett, T. J.; Yost, R. A., Comparison of blood plasma sample preparation methods for combined LC–MS lipidomics and metabolomics. Journal of Chromatography B 2015, 1002, 260-266.

Peterson, A. C.; Balloon, A. J.; Westphall, M. S.; Coon, J. J., Development of a GC/Quadrupole-Orbitrap Mass Spectrometer, Part II: New Approaches for Discovery Metabolomics. Analytical Chemistry 2014, 86 (20), 10044-10051.

Peterson, A. C.; Hauschild, J.-P.; Quarmby, S. T.; Krumwiede, D.; Lange, O.; Lemke, R. A. S.; Grosse-Coosmann, F.; Horning, S.; Donohue, T. J.; Westphall, M. S.; Coon, J. J.; Griep-Raming, J., Development of a GC/Quadrupole-Orbitrap Mass Spectrometer, Part I: Design and Characterization. Analytical Chemistry 2014, 86 (20), 10036-10043.

Petherick, A., Development: Mother's milk: A rich opportunity. Nature 2010, 468 (7327), S5-S7

Petro, E. M. L.; Leroy, J. L. M. R.; Covaci, A.; Fransen, E.; De Neubourg, D.; Dirtu, A. C.; De Pauw, I.; Bols, P. E. J., Endocrine-disrupting chemicals in human follicular fluid impair in vitro oocyte developmental competence. Human Reproduction 2012, 27 (4), 1025- 1033.

Pico, Y.; Alfarhan, A. H.; Barcelo, D., How recent innovations in gas chromatography- mass spectrometry have improved pesticide residue determination: An alternative technique to be in your radar. TrAC Trends in Analytical Chemistry 2020, 122, 115720.

Pluskal, T.; Castillo, S.; Villar-Briones, A.; Orešič, M., MZmine 2: modular framework for processing, visualizing, and analyzing mass spectrometry-based molecular profile data. BMC bioinformatics 2010, 11 (1), 395.

Pourchet, M.; Debrauwer, L.; Klanova, J.; Price, E. J.; Covaci, A.; Caballero-Casero, N.; Oberacher, H.; Lamoree, M.; Damont, A.; Fenaille, F.; Vlaanderen, J.; Meijer, J.; Krauss, M.; Sarigiannis, D.; Barouki, R.; Le Bizec, B.; Antignac, J.-P., Suspect and non-targeted screening of chemicals of emerging concern for human biomonitoring, environmental health studies and support to risk assessment: From promises to challenges and harmonisation issues. Environment International 2020, 139, 105545.

- 238/296 - List of bibliographical references

Rappaport, S. M.; Barupal, D. K.; Wishart, D.; Vineis, P.; Scalbert, A., The blood exposome and its role in discovering causes of disease. Environmental health perspectives 2014, 122 (8), 769-774.

Rappaport, S. M.; Smith, M. T., Environment and Disease Risks. Science 2010, 330 (6003), 460-461.

Ren, X. M.; Guo, L.-H., Assessment of the Binding of Hydroxylated Polybrominated Diphenyl Ethers to Thyroid Hormone Transport Proteins Using a Site-Specific Fluorescence Probe. Environmental Science & Technology 2012, 46 (8), 4633-4640.

Ribas-Fitó, N.; Ramón, R.; Ballester, F.; Grimalt, J.; Marco, A.; Olea, N.; Posada, M.; Rebagliato, M.; Tardón, A.; Torrent, M.; Sunyer, J.; Group, o. b. o. t. I. S., Child health and the environment: the INMA Spanish Study. Paediatric and Perinatal Epidemiology 2006, 20 (5), 403-410.

Rochat, B., Proposed confidence scale and ID score in the identification of known- unknown compounds using high resolution MS data. Journal of The American Society for Mass Spectrometry 2017, 28 (4), 709-23.

Saito, K.; Sjödin, A.; Sandau, C. D.; Davis, M. D.; Nakazawa, H.; Matsuki, Y.; Patterson Jr, D. G., Development of a accelerated solvent extraction and gel permeation chromatography analytical method for measuring persistent organohalogen compounds in adipose and organ tissue analysis. Chemosphere 2004, 57 (5), 373-381.

Samanipour, S.; Baz-Lomba, J. A.; Reid, M. J.; Ciceri, E.; Rowland, S.; Nilsson, P.; Thomas, K. V., Assessing sample extraction efficiencies for the analysis of complex unresolved mixtures of organic pollutants: A comprehensive non-target approach. Analytica chimica acta 2018, 1025, 92-98.

Saradha, B.; Mathur, P. P., Effect of environmental contaminants on male reproduction. Environmental Toxicology and Pharmacology 2006, 21 (1), 34-41.

Sauve, S.; Desrosiers, M., A review of what is an emerging contaminant. Chem Cent J 2014, 8 (1), 15.

- 239/296 - List of bibliographical references

Schymanski, E. L.; Jeon, J.; Gulde, R.; Fenner, K.; Ruff, M.; Singer, H. P.; Hollender, J., Identifying Small Molecules via High Resolution Mass Spectrometry: Communicating Confidence. Environmental Science & Technology 2014, 48 (4), 2097-2098.

Schymanski, E. L.; Williams, A. J., Open Science for Identifying “Known Unknown” Chemicals. Environmental Science & Technology 2017, 51 (10), 5357-5359.

Sexton, K.; Callahan, M. A.; Bryan, E. F., Estimating exposure and dose to characterize health risks: the role of human tissue monitoring in exposure assessment. Environmental Health Perspectives 1995, 103 (suppl 3), 13-29.

Simon, E.; Lamoree, M. H.; Hamers, T.; de Boer, J., Challenges in effect-directed analysis with a focus on biological samples. TrAC Trends in Analytical Chemistry 2015, 67, 179-191.

Simon, E.; van Velzen, M.; Brandsma, S. H.; Lie, E.; Løken, K.; de Boer, J.; Bytingsvik, J.; Jenssen, B. M.; Aars, J.; Hamers, T.; Lamoree, M. H., Effect-Directed Analysis to Explore the Polar Bear Exposome: Identification of Thyroid Hormone Disrupting Compounds in Plasma. Environmental Science & Technology 2013, 47 (15), 8902-8912.

Sleno, L., The use of mass defect in modern mass spectrometry. Journal of Mass Spectrometry 2012, 47 (2), 226-236

Smolders, R.; Schramm, K.-W.; Stenius, U.; Grellier, J.; Kahn, A.; Trnovec, T.; Sram, R.; Schoeters, G., A Review on the Practical Application of Human Biomonitoring in Integrated Environmental Health Impact Assessment. Journal of Toxicology and Environmental Health, Part B 2009, 12 (2), 107-123.

Sobus, J. R.; Grossman, J. N.; Chao, A.; Singh, R.; Williams, A. J.; Grulke, C. M.; Richard, A. M.; Newton, S. R.; McEachran, A. D.; Ulrich, E. M., Using prepared mixtures of ToxCast chemicals to evaluate non-targeted analysis (NTA) method performance. Analytical and Bioanalytical Chemistry 2019, 411 (4), 835-851.

Spicer, R.; Salek, R. M.; Moreno, P.; Cañueto, D.; Steinbeck, C., Navigating freely- available software tools for metabolomics analysis. Metabolomics 2017, 13 (9), 106.

- 240/296 - List of bibliographical references

Stanstrup, J.; Broeckling, C. D.; Helmus, R.; Hoffmann, N.; Mathé, E.; Naake, T.; Nicolotti, L.; Peters, K.; Rainer, J.; Salek, R. M.; Schulze, T.; Schymanski, E. L.; Stravs, M. A.; Thévenot, E. A.; Treutler, H.; Weber, R. J.; Willighagen, E.; Witting, M.; Neumann, S., The metaRbolomics Toolbox in Bioconductor and beyond. Metabolites 2019, 9 (10), 200.

Steckling, N.; Gotti, A.; Bose-O’Reilly, S.; Chapizanis, D.; Costopoulou, D.; De Vocht, F.; Garí, M.; Grimalt, J. O.; Heath, E.; Hiscock, R.; Jagodic, M.; Karakitsios, S. P.; Kedikoglou, K.; Kosjek, T.; Leondiadis, L.; Maggos, T.; Mazej, D.; Polańska, K.; Povey, A.; Rovira, J.; Schoierer, J.; Schuhmacher, M.; Špirić, Z.; Stajnko, A.; Stierum, R.; Tratnik, J. S.; Vassiliadou, I.; Annesi-Maesano, I.; Horvat, M.; Sarigiannis, D. A., Biomarkers of exposure in environment- wide association studies – Opportunities to decode the exposome using human biomonitoring data. Environmental Research 2018, 164, 597-624

Stettin, D.; Poulin, R. X.; Pohnert, G., Metabolomics Benefits from Orbitrap GC–MS— Comparison of Low- and High-Resolution GC–MS. Metabolites 2020, 10 (4), 143.

SusDat, Merged NORMAN Suspect List: SusDat, 2012 https://www.norman- network.com/?q=node/236; Norman Network.

Taguchi, V. Y.; Nieckarz, R. J.; Clement, R. E.; Krolik, S.; Williams, R., Dioxin analysis by gas chromatography-Fourier transform ion cyclotron resonance mass spectrometry (GC-FTICRMS). Journal of the American Society for Mass Spectrometry 2010, 21 (11), 1918- 1921.

Tautenhahn, R.; Patti, G. J.; Rinehart, D.; Siuzdak, G., XCMS Online: a web-based platform to process untargeted metabolomic data. Analytical chemistry 2012, 84 (11), 5035- 5039.

Tsugawa, H.; Cajka, T.; Kind, T.; Ma, Y.; Higgins, B.; Ikeda, K.; Kanazawa, M.; VanderGheynst, J.; Fiehn, O.; Arita, M., MS-DIAL: data-independent MS/MS deconvolution for comprehensive metabolome analysis. Nature methods 2015, 12 (6), 523.

Tugizimana, F.; Steenkamp, P.; Piater, L.; Dubery, I., A conversation on data mining strategies in LC-MS untargeted metabolomics: Pre-processing and pre-treatment steps. Metabolites 2016, 6 (4), 40.

- 241/296 - List of bibliographical references

Turyk, M. E.; Anderson, H. A.; Persky, V. W., Relationships of thyroid hormones with polychlorinated biphenyls, dioxins, furans, and DDE in adults. Environmental health perspectives 2007, 115 (8), 1197-1203.

Ulmer, C. Z.; Jones, C. M.; Yost, R. A.; Garrett, T. J.; Bowden, J. A., Optimization of Folch, Bligh-Dyer, and Matyash sample-to-extraction solvent ratios for human plasma-based lipidomics studies. Analytica chimica acta 2018, 1037, 351-357.

Ulrich, E. M.; Sobus, J. R.; Grulke, C. M.; Richard, A. M.; Newton, S. R.; Strynar, M. J.; Mansouri, K.; Williams, A. J., EPA’s non-targeted analysis collaborative trial (ENTACT): genesis, design, and initial findings. Analytical and Bioanalytical Chemistry 2019, 411 (4), 853- 866.

UNEP Stockholm Convention on Persistant Organic pollutants (POPs). http://chm.pops.int/Home/tabid/2121/Default.aspx.

US-EPA EcoBox Tools by exposure Pathways - Exposure Pathways in Ecological Risk Assessment (ERA). https://www.epa.gov/ecobox/epa-ecobox-tools-exposure-pathways- exposure-pathways-era.

van Den Dool, H.; Dec. Kratz, P., A generalization of the retention index system including linear temperature programmed gas—liquid partition chromatography. Journal of Chromatography A 1963, 11, 463-471.

Vazquez-Roig, P.; Picó, Y., Pressurized liquid extraction of organic contaminants in environmental and food samples. TrAC Trends in Analytical Chemistry 2015, 71, 55-64.

Vermeulen, R.; Schymanski, E. L.; Barabási, A.-L.; Miller, G. W., The exposome and health: Where chemistry meets biology. Science 2020, 367 (6476), 392-396.

Vizcaino, E.; Grimalt, J. O.; Fernández-Somoano, A.; Tardon, A., Transport of persistent organic pollutants across the human placenta. Environment International 2014, 65, 107-115.

Warth, B.; Spangler, S.; Fang, M.; Johnson, C. H.; Forsberg, E. M.; Granados, A.; Martin, R. L.; Domingo-Almenara, X.; Huan, T.; Rinehart, D.; Montenegro-Burke, J. R.; Hilmers, B.; Aisporna, A.; Hoang, L. T.; Uritboonthai, W.; Benton, H. P.; Richardson, S. D.;

- 242/296 - List of bibliographical references

Williams, A. J.; Siuzdak, G., Exposome-Scale Investigations Guided by Global Metabolomics, Pathway Analysis, and Cognitive Computing. Analytical Chemistry 2017, 89 (21), 11505- 11513.

WHO, Global assessment of the state-of-the-science of endocrine disruptors - part 1 "Executive summary"; WHO/PCS/EDC/02.2; 2002.

WHO, State of the science of endocrine disrupting chemicals, an assessment of the state of the science of endocrine disruptors prepared by a group of experts for the United Nations Environment Programme (UNEP) and WHO, 2012.

WHO/IPCS Environmental Health Criteria No 155 - Biomarkers and risk assessment: concepts and principles; 1993.

Wild, C. P., Complementing the Genome with an “Exposome”: The Outstanding Challenge of Environmental Exposure Measurement in Molecular Epidemiology. Cancer Epidemiology Biomarkers & Prevention 2005, 14 (8), 1847-1850.

Wold, S.; Kettaneh-Wold, N.; Skagerberg, B., Nonlinear PLS modeling. Chemometrics and Intelligent Laboratory Systems 1989, 7 (1), 53-65.

Yusa, V.; Millet, M.; Coscolla, C.; Roca, M., Analytical methods for human biomonitoring of pesticides. A review. Analytica Chimica Acta 2015, 891, 15-31.

Yusa, V.; Ye, X.; Calafat, A. M., Methods for the determination of biomarkers of exposure to emerging pollutants in human specimens. TrAC Trends in Analytical Chemistry 2012, 38, 129-142.

Zhang, X.; Romm, M.; Zheng, X.; Zink, E. M.; Kim, Y.-M.; Burnum-Johnson, K. E.; Orton, D. J.; Apffel, A.; Ibrahim, Y. M.; Monroe, M. E.; Moore, R. J.; Smith, J. N.; Ma, J.; Renslow, R. S.; Thomas, D. G.; Blackwell, A. E.; Swinford, G.; Sausen, J.; Kurulugama, R. T.; Eno, N.; Darland, E.; Stafford, G.; Fjeldsted, J.; Metz, T. O.; Teeguarden, J. G.; Smith, R. D.; Baker, E. S., SPE-IMS-MS: an automated platform for sub-sixty second surveillance of endogenous metabolites and xenobiotics in biofluids. Clinical Mass Spectrometry 2016, 2, 1- 10.

- 243/296 - List of bibliographical references

Zhao, L.; Lucas, D.; Long, D.; Richter, B.; Stevens, J., Multi-class multi-residue analysis of veterinary drugs in meat using enhanced matrix removal lipid cleanup and liquid chromatography-tandem mass spectrometry. Journal of Chromatography A 2018, 1549, 14-24.

Zhao, L.; Szakas, T.; Churley, M.; Lucas, D., Multi-class multi-residue analysis of pesticides in edible oils by gas chromatography-tandem mass spectrometry using liquid-liquid extraction and enhanced matrix removal lipid cartridge cleanup. Journal of Chromatography A 2019, 1584, 1-12.

- 244/296 -

LIST OF ABBREVIATIONS

- 245/296 -

List of abbreviations

GENERALITIES

HBM: Human Biomonitoring APCI: Atmospheric Pressure Chemical Conisation HBM4EU: Human Biomonitoring for Europe APPI: Atmospheric Pressure Photoionisation HILIC: Hydrophilic Interaction Liquid Chromatography AU: Arbitrary Unit HMDB: Human Metabolome DataBase BFR: Brominated Flame Retardants HRMS: High Resolution Mass Captiva EMR-Lipid®: Enhanced Matrix Spectrometry Removal of lipids INRAE: Institut National de Recherche CDC: Centers for Disease Control and pour l’Agriculture, l’Alimentation et Prevention l’Environnement (French National CEC: Chemical emerging compounds Research Institute for Agriculture, Food and Environment) DDA: Data Dependent Acquisition IPCS: International Programme on DIA: Data Independent Acquisition Chemical Safety

ECHA: European Chemicals Agency IUPAC: International Union of Pure and Applied Chemistry EDA: Effect Direct Analysis LABERCA: Laboratoire d’Etude des EDC: Endocrine Disrupting Chemicals Résidus et Contaminants dans les Aliments

EEA: European Environment Agency LC: Liquid Chromatography

EI: Electronic Impact LLE: Liquid-Liquid Extraction

EIC: Extracted Ion Chromatogram LOD: Limit of Detection

ESI: Electrospray Ionisation LRMS: Low Resolution Mass Spectrometry FAME: Fatty Acid Methyl Esters m/z: mass to charge ratio FAO: Food and Agriculture Organisation MAE: Microwave Assisted Extraction FT-ICR: Fourier Transform Ion Cyclotron Resonance MD plot: Mass Defect plot

GC: Gas Phase Chromatography METLIN

GPC: Gel Permeation Chromatography MS/MS: tandem mass spectrometry

- 247/296 - List of abbreviations

MS: Mass Spectrometry UNEP: United Nations Environment Programme NCE: Normalised Collision Energy US EPA: Unitated State Environmental NIST: Nation Institute of Standards and protection agency Technology VITO: Flemish Center of Expertise on NMR: Nuclear Magnetic Resonance Environment and Health

NTS: Non-Targeted Screening WHO: World Health Organisation

PLE: Pressurised Liquid Extraction WP: Work-Packages

POP: Persistent Organic Pollutants QA/QC: Quality Assurance/Quality Control q-MS: Quadrupole

QqQ: triple quadrupole

QuEChERS: Quick, Easy, Cheap, Effective, Rugged and Safe

Ri: Retention index

ROI: Region Of Interest

RSD: Residual Standard Deviation rT: retention time

SEC: Size Exclusion Chromatography

SFC: Supercritical fluid chromatography

SLE: Solid-Liquid Extraction

SPE: Solid Phase Extraction (d-SPE dispersive-SPE)

SRM: Selected Reaction Monitoring

TIC: Total Ion Current

ToF: Time-of-Flight

UBA: Umweltbundesamt

- 248/296 - List of abbreviations

CHEMICALS

AcN: Acetonitrile

Anti-DP: Anti-dechlorane plus

DDT: Dichlorodiphenyltrichloroethane

HBBz: Hexabromobenzene

HBCDD: Hexabromocyclododecane

HCB: Hexachlorobenzene

HCH: Hexachlorocyclohexane

MTBE: Methyl tert-butyl ether

P,p,’-DDE: Dichlorodiphenyltrichloroethane

PBDE Polybrominated diphenyl ethers

PCB: Polychloro Biphenyls

PFOS: Perfluoroctane sulfonate

TBBPA: Tetrabromobisphenol A

OH-BDE 137: 6-hydroxy-2,2’,3,4,4’,5-hexabromodiphenyl ether

2,4,3,5-tetraBP: 2,3,4,5-tetrabromophenol

2,4,3,5-tetraCP: 2,3,4,5-tetrachlorophenol

BTBPE: 1,2-bis(2,4,6-tribromophenoxy) ethane

TCPy: 3,5,6-trichloro-2-pyridinol

BDCIPP: bis(1,3-dichloro-2-propyl) phosphate

2,4-DBP: 2,4-dibromophenol

2,4-DCP: 2,4-dichlorophenol

PBDE 153: 2,2’,4,4’,5,5’-hexabromodiphenyl ether

- 249/296 -

LIST OF FIGURES

251/296 -

List of figures

FIGURE 1.1: ILLUSTRATION OF THE ENVIRONMENT-FOOD-HUMAN CONTINUUM AND EXTERNAL HUMAN EXPOSURE PATHWAYS.

ADAPTED FROM US EPA ECOBOX TOOLS BY EXPOSURE PATHWAYS AND VERMEULEN ET AL., 2020...... 38

FIGURE 1.2: GENERIC VIEW OF CONTAMINANT ABSORPTION, DISTRIBUTION AND EXCRETION WITH DIFFERENT COLLECTABLE MATRICES

IN BLUE. ADAPTED FROM SEXTON ET AL., 1995 AND NEEDHAM ET AL., 2005...... 40

FIGURE 1.3: ADDRESSING THE EXPOSOME THROUGH A TRIAL COLLABORATION BETWEEN SCIENTISTS, CHEMICAL RISK ASSESSORS AND

RISK MANAGERS OVER HUMAN BIOMONITORING PROGRAMMES...... 42

FIGURE 1.4: HBM4EU PROJECT SKELETON...... 44

FIGURE 1.5: CONCEPTUAL VIEW OF THE HUMAN CHEMICAL EXPOSOME, RELATED METHODOLOGICAL APPROACHES AND ASSOCIATED

PURPOSES ...... 46

FIGURE 1.6: CHEMICAL SPACE DETECTABLE BY APPLYING A GIVEN CHROMATOGRAPHY COUPLED TO MASS SPECTROMETRY BASED ON

METHODOLOGICAL SCREENING WORKFLOW...... 48

FIGURE 1.7: PRINCIPLE OF CHROMATOGRAPHY COUPLED TO MASS SPECTROMETRY...... 49

FIGURE 1.8: COMPLEMENTARITY OF LC-MS AND GC-MS BASED SYSTEMS IN TERMS OF DETECTABILITY OF COMPOUNDS WITH

VARIOUS VOLATILITY AND POLARITY. ADAPTED FROM BRACK ET AL 2016...... 56

FIGURE 1.9: SUMMARISED CONCEPTUAL COMPARISON OF THE ANALYTICAL WORKFLOWS TYPICALLY APPLIED FOR CONVENTIONAL

TARGETED (LEFT) VERSUS NON-TARGETED (RIGHT) METHODS (A), AND OF THE RESULTING GLOBAL PERFORMANCE EXPECTED

WITH BOTH APPROACHES (B). RADAR CHART AXIS SCALE IS IN ARBITRARY UNITS WITH 0 CORRESPOND TO LOW PERFORMANCE

AND 100 A HIGH PERFORMANCE...... 61

FIGURE 1.10: PRINCIPLE OF EFFECT-DIRECTED ANALYSIS COMBINING CHEMICAL AND BIOLOGICAL APPROACH. AFTER THE SAMPLE

TREATMENT, EXTRACT IS SEPARATED BY LIQUID CHROMATOGRAPHY AND THE POST-COLUMN FLOW IS SPLIT IN TWO. ONE PART IS

DEDICATED TO HRMS ANALYSIS IN FULL-SCAN MODE AND THE OTHER PART IS FRACTIONATED IN A 96 WELLS PLATE. THE

BIOLOGICAL ACTIVITY OF EACH WELL IS MEASURED WITH THE BIOLOGICAL TEST OF INTEREST AND THE BIOLOGICAL RESPONSE IS

INTERPRETED WITH REGARDS TO MS DATA...... 62

FIGURE 1.11: FLUORESCENCE RESPONSE OF THE T4-FITC COMPLEX (RED TRIANGLE) BOUND TO TTR PROTEIN (BLUE HEXAGON) AS A

FUNCTION OF THE THYROID HORMONE SYSTEM DISRUPTORS (YELLOW HEXAGON) CONCENTRATION...... 63

FIGURE 1.12: IDENTIFICATION CONFIDENCE LEVEL PROPOSED BY SCHYMANSKI ET AL., 2014 FOR HRMS DATA...... 66

FIGURE 1.13: SUMMARISED CONCEPTUAL COMPARISON OF THE QA/QC CURRENT STATE OF DEVELOPMENT TYPICALLY OBSERVED FOR

CONVENTIONAL TARGETED (LEFT) VERSUS NON-TARGETED (RIGHT) METHODS...... 68

FIGURE 2.1: VENN DIAGRAM ILLUSTRATING THE LC-ESI(+/-)-HRMS AND GC-EI-HRMS COMPLEMENTARITY TO DETECT THE

SELECTED RANGE OF STANDARDS OF THE QA/QC MIX 2 USED TO GUIDE THE PRESENT METHODOLOGICAL DEVELOPMENT (30

TEST REFERENCE COMPOUNDS LISTED IN TABLE 2.2 AND APPENDIX 2) (A). ILLUSTRATION OF THEIR DIFFERENT

PHYSICOCHEMICAL PROPERTIES: MONOISOTOPIC MASS AND LOG P (B). BLUE, RED AND BLACK TRIANGLES ARE COMPOUNDS

DETECTED IN LC-ESI(-)-HRMS, LC-ESI(+)-HRMS AND GC-EI-HRMS, RESPECTIVELY. A COMBINATION OF COLOUR

INDICATES DUAL MODE OF DETECTION. YELLOW (NO. 16) DOT IS THE COMPOUND DETECTED BY THREE MODES...... 79

FIGURE 2.2: EXTRACTED ION CHROMATOGRAMS BY LC-ESI(-)-HRMS FOR 6 TRIBROMOPHENOL ISOMERS (LEFT) AND Α- AND Γ-HBCD

(RIGHT), OBTAINED AFTER CHROMATOGRAPHIC SEPARATION ON FOUR STATIONARY PHASES NAMELY HYPERSIL GOLD, ACQUITY

BEH C18, HYPERSIL GOLD PFP AND ACQUITY CSH C18 (TOP TO BOTTOM)...... 81

- 253/296 - List of figures

FIGURE 2.3: TOTAL ION CURRENTS OBTAINED BY LC-ESI(-)-HRMS FOR PLACENTA SAMPLE EXTRACT ELUTED ON HYPERSIL GOLD

COLUMN WITH WATER AND ACETONITRILE SUPPLEMENTED WITH AMMONIUM ACETATE 10 MM, AMMONIUM FLUORIDE 0.2

AND 1 MM AND ACETIC ACID 0.1 % (TOP TO BOTTOM)...... 82

FIGURE 2.4: EXTRACTION ION CHROMATOGRAMS OBTAINED FOR 3 TETRACHLOROPHENOL ISOMERS AFTER SEPARATION ON HYPERSIL

GOLD COLUMN WITH WATER AND ACETONITRILE SUPPLEMENTED WITH AMMONIUM ACETATE 10 MM (TOP) AND ACETIC ACID

0.1 % (BOTTOM) ...... 83

FIGURE 2.5: EXTRACTED ION CHROMATOGRAMS OBTAINED FOR STANDARDS OF THE QA/QC MIX 2 AFTER SEPARATION ON HYPERSIL

GOLD COLUMN WITH THE THREE TESTED ELUTIONS GRADIENTS G1, G2 AND G3...... 84

FIGURE 2.6: LC-ESI(-)-HRMS TOTAL ION CURRENTS (LEFT) OBTAINED FOR A PLACENTA SAMPLE EXTRACT AND EXTRACTED ION

CHROMATOGRAMS IN THE SAME SAMPLE OF TBBPA AND Α-HBCD (RIGHT) AT THE BEGINNING AND THE END OF THE BATCH

WITH INSUFFICIENTLY OPTIMISED COLUMN FLUSH...... 85

FIGURE 2.7: OPTIMISATION OF THE ION SOURCE TEMPERATURE. DETECTED INTENSITY FOR COMPOUNDS OF QA/QC MIX 2 IN LC-

ESI(-) (LEFT) AND LC-ESI(+) (RIGHT) FOR THREE TEMPERATURE OF 150, 250 AND 350 °C WITH S-LENS OF 50 AU...... 87

FIGURE 2.8: OPTIMISATION OF ION SOURCE S-LENS VALUE. DETECTED INTENSITY FOR COMPOUNDS OF QA/QC MIX 2 IN LC-ESI(-)

(LEFT) AND LC-ESI(+) (RIGHT) FOR A SOURCE TEMPERATURE OF 350 °C AND FOUR S-LENS VALUES OF 30, 50, 70 AND 100 AU...... 88

FIGURE 2.9: GC-EI-HRMS EXTRACTED ION CHROMATOGRAM OBTAINED FOR ACETOCHLOR, METOLACHLOR AND P-TBX INJECTED IN

TOLUENE (LEFT) AND HEXANE (RIGHT)...... 91

FIGURE 2.10: TOTAL ION CURRENTS OBTAINED FOR EMPTY VIAL CLOSED BY SILICON/PTFE (BLACK) OR NATURAL RUBBER (RED)

SEPTUM OR ALUMINIUM FOIL (BLUE) TO COMPARE EXTERNAL CONTAMINATION RELEASED BY SEPTUM (LEFT). TOTAL ION

CURRENTS OBTAINED FOR VIAL CONTAINING HEXANE CLOSED BY SILICON/PTFE SEPTUM INJECTED TWICE WITH COMPARISON TO

VIAL CONTAINING HEXANE CLOSED BY ALUMINIUM FOIL (RIGHT). ANALYSES WERE PERFORMED IN GC-EI-HRMS...... 92

FIGURE 2.11: GC-EI-HRMS EXTRACTED ION CHROMATOGRAMS OBTAINED FOR COMPOUNDS OF THE QA/QC MIX 2 ELUTED WITH

THE FOLLOWING GRADIENT OF TEMPERATURE: 60 °C FOR 2 MIN, INCREASED TO 130 °C AT 10 °C/MIN, THEN TO 250 °C AT 5

°C/MIN AND TO 320 °C AT 10 °C/MIN AND HELD AT 320 °C FOR 10 MIN...... 93

FIGURE 2.12: THEORETICAL ISOTOPIC PATTERNS RESULTING FROM 1 TO 10 CHLORINATED (LEFT) AND BROMINATED (RIGHT) ATOMS IN

MASS SPECTROMETRY...... 101

FIGURE 2.13: MD PLOT OBTAINED FROM HALOSEEKER V.1.0 WITH CLUSTERS OF MONO- TO PENTA-CHLOROPHENOLS (IN

RECTANGLES) AND MONO- TO PENTA-BROMOPHENOLS (IN CIRCLES) HORIZONTALLY ALIGNED...... 102

FIGURE 2.14: SYNOPTIC VIEW OF HALOSEEKER DATA PROCESSING WORKFLOW IN THE BACK END...... 103

FIGURE 2.15: SYNTHETIC VIEW OF A H/CL-SCALE MD PLOT (LEFT) AND DETAILS OF TWO CL-CONTAINING HOMOLOGUE CLUSTERS

(RIGHT). EACH DOT REPRESENTS A GIVEN FEATURES CHARACTERISED BY A GIVEN M/Z, RETENTION TIME AND INTENSITY VALUES;

GROUP OF PAIRED FEATURES, OR ISOTOPOLOGUE, IS DEFINED AS A CLUSTER...... 104

FIGURE 2.16: MD PLOT OBTAINED BY THE HALOSEEKER V1.0 SOFTWARE AFTER THE PROCESSING OF TYPICAL LC-ESI(-)-HRMS DATA

FILE ACQUIRED FROM A SOLVENT SAMPLE CONTAINING STANDARDS OF THE QA/QC MIX 2...... 108

FIGURE 2.17: SUMMARY OF THE DEVELOPED STRATEGY TO PROCESS NON-TARGETED LC-HRMS DATA. *: STEP PERFORMED

MANUALLY...... 109

- 254/296 - List of figures

FIGURE 2.18: ITERATIVE-DDA STRATEGY USED IN LC-HRMS TO GENERATE MS/MS SPECTRA. FOR THE PRESENT EXAMPLE, A SINGLE

SAMPLE IS INJECTED THREE TIME WITHIN TOP-5 DDA CONDITION. AN EXCLUSION LIST IS CREATED AFTER EACH INJECTION AND

MERGED TO THE PREVIOUS EXCLUSION LIST IN ORDER TO FRAGMENTED LESS AND LESS INTENSE IONS...... 112

FIGURE 2.19: EXAMPLES OF REFERENCE MS/MS SPECTRA GENERATED FOR FIPRONIL (20 NG µL-1) AT NCE 10, 20, 30, 40, 50 AND

60 ACQUIRED IN A SINGLE ACQUISITION TO INPUT INTO THE ANNOTATION DATABASE DEVELOPED FOR MARKERS OF CHEMICAL

EXPOSURE WITHIN THE HBM4EU PROJECT...... 113

FIGURE 3.1: LOGKOW-VALUES AS A FUNCTION OF MONOISOTOPIC MASSES FOR COMPOUNDS INCLUDED IN THE MERGED NORMAN

SUSPECT LIST INVENTORYING KNOWN EMERGING CHEMICALS IN ENVIRONMENTAL MATRICES (SUSDAT, 2012)...... 122

FIGURE 3.2: OVERLAP OF TOTAL ION CURRENTS OBTAINED FOR A HUMAN ADIPOSE TISSUE (BLUE) AND A PROCEDURAL BLANK (BLACK)

SAMPLES PURIFIED WITH AN ACIDIC HYDROLYSIS AND ANALYSED BY LC-ESI(-)-HRMS...... 125

FIGURE 3.3: ILLUSTRATION OF THE TYPICAL EXPERIMENT SET UP TO CHARACTERISE THE ION SUPPRESSION PHENOMENON BY LC-

HRMS. A PURE STANDARD OF A GIVEN COMPOUND IS POST-COLUMN INTRODUCED IN THE DETECTION SYSTEM TOGETHER WITH

THE PREPARED SAMPLE EXTRACT. THE FLUCTUATION ON THE EXTRACTION ION CHROMATOGRAM CHARACTERISE MATRIX EFFECT.

ADAPTED FROM ANTIGNAC ET AL., 2005...... 126

FIGURE 3.4: OVERLAP OF THE LC-ESI(-)-HRMS EXTRACTED ION CHROMATOGRAMS OF HYDROXY-TRICHLORODIPHENYL ETHER

(HTCDE), HYDROXY-TRIBROMODIPHENYL ETHER (HTBDE) AND TETRABROMO BISPHENOL A (TBBPA) OBSERVED FOR TYPICAL

ADIPOSE TISSUE (BLUE), PROCEDURAL BLANK (RED) AND MOBILE PHASE (BLACK) SAMPLES (FROM TOP TO BOTTOM) IN PRESENCE

OF A POST-COLUMN INTRODUCTION OF A SOLUTION OF HTCDE, HTBDE AND TBBPA. LAST CHROMATOGRAMS (BOTTOM)

SHOW THE OBSERVED SIGNALS FOR THE THREE REFERENCE COMPOUNDS INJECTED AS THE MIX OF STANDARDS WITHOUT POST-

COLUMN INFUSION...... 127

FIGURE 3.5: OPTIMISATION OF THE GPC FRACTIONATION PURIFICATION. GRAVIMETRIC DETERMINATION (A) AND UV SIGNAL AT 254

NM (B) OF THE FAT CONTENT OF ADIPOSE TISSUE FRACTIONATED BY 2 MIN SEGMENTS. LC-ESI(-)-HRMS ANALYSIS OF

FRACTIONS COLLECTED BY GPC FROM A SAMPLE OF CL/BR-PHENOLIC COMPOUNDS OF STANDARD MIX 1, FRACTIONATED BY 2

MIN SEGMENTS. EXTRACTED ION CHROMATOGRAMS FOR EACH STANDARD IN EACH FRACTION WERE INTEGRATED AND THE PEAK

AREA IS REPORTED IN C...... 130

FIGURE 3.6: OVERLAP OF LC-ESI(-)-HRMS TOTAL ION CURRENTS OBTAINED FOR ADIPOSE TISSUE (BLUE) AND PROCEDURAL BLANK

(BLACK) SAMPLES PURIFIED ACCORDING TO THE DEVELOPED GPC PROCEDURE...... 131

FIGURE 3.7: MD PLOT OBTAINED WITH HALOSEEKER V1.0 SOFTWARE FOLLOWING THE DATA PROCESSING OF A PROCEDURAL BLANK

SAMPLE PURIFIED BY GPC AND ANALYSED BY LC-ESI(-)-HRMS (TOP). THE PARTICULAR SIGNAL OF TBBPA IS INDICATED IN THE

BLACK CIRCLE. EXPERIMENTALLY OBSERVED MASS SPECTRA (BOTTOM LEFT, DOWN RED TRACES) AND EXTRACTED ION

CHROMATOGRAM (BOTTOM RIGHT) CORRESPONDING TO THIS SIGNAL...... 133

FIGURE 3.8: OVERLAPPED LC-ESI(-)-HRMS TOTAL ION CURRENTS (TOP) OBSERVED FOR PROCEDURAL BLANK SAMPLES PREPARED

WITH GPC (BLUE), ROTARY EVAPORATOR (RED), LEFT OVERNIGHT UNDER THE FUME HOOD (GREEN) AND MOBILE PHASE

(BLACK). EXTRACTED ION CHROMATOGRAM OF TBBPA OBSERVED IN PROCEDURAL BLANK SAMPLE PREPARED WITH GPC

(BOTTOM LEFT) AND THE CORRESPONDING MASS SPECTRUM ANNOTATED WITH MASS DEVIATION IN BRACKETS (BOTTOM

RIGHT)...... 135

FIGURE 3.9: BLIGH AND DYER PROTOCOL APPLIED TO 100 MG PLACENTA SAMPLES ASSISTED BY PRECELLYS® HOMOGENISER DEVICE...... 138

- 255/296 - List of figures

FIGURE 3.10: TOTAL ION CURRENTS OBTAINED FOR PLACENTA SAMPLES EXTRACTED WITH THE BLIGH AND DYER APPROACH RESULTING

IN NONPOLAR (CHLOROFORM, ON TOP) AND POLAR (WATER/METHANOL, ON BOTTOM) FRACTIONS. EXTRACTS WERE ANALYSED

BY LC-ESI(-)-HRMS ON THE SCAN RANGE 100-1 000 DA (BLACK) AND 1 000-2 000 DA (RED) AND BY LC-ESI(+)-HRMS

ON THE SCAN RANGE 100-1 000 DA (BLUE) AND 1 000-2 000 DA (GREEN)...... 139

FIGURE 3.11: RESULTS OF PARTITIONING (LC-ESI(-)-HRMS DETECTION) OBSERVED FOR THE DIFFERENT CL/BR-PHENOLIC

COMPOUNDS OF THE STANDARD MIX 1 SELECTED AS MARKERS, SPIKED IN A HUMAN PLACENTA SAMPLE, BETWEEN THE

NONPOLAR (CHLOROFORM IN RED) AND THE POLAR (METHANOL/WATER IN BLUE) FRACTION AFTER THE BLIGH AND DYER

PROTOCOL...... 141

FIGURE 3.12: ESTIMATED RECOVERY FOR THE DIFFERENT CL/BR-PHENOLIC COMPOUNDS OF THE STANDARD MIX 1, TESTED AS

MARKERS SPIKED IN PLACENTA PREPARED WITH THE BLIGH AND DYER PROTOCOL...... 142

FIGURE 3.13: CONCENTRATION (NG G -1 LIPID) OF ORGANIC CONTAMINANTS DETECTED IN PLACENTA SAMPLES PREPARED WITH THE

TARGETED METHOD DEVELOPED BY LABERCA AND ANALYSED IN GC-EI-MAGNETIC SECTOR, LC-ESI-QQQ, LC-ESI-ORBITRAP,

APGC-QQQ...... 145

FIGURE 3.14: INFLUENCE OF THE SAMPLE AMOUNT CONSIDERED FOR ANALYSIS ON COMPOUNDS DETECTION. OVERLAP OF LC-ESI(-)-

HRMS TOTAL ION CURRENTS (TOP) OBTAINED FOR A PROCEDURAL BLANK (BLACK) AND MILK SAMPLES OF 5 ML (GREEN), 500

µL (BLUE) AND 50 µL (RED). OVERLAP OF EXTRACTED ION CHROMATOGRAMS (BOTTOM) OF 13C-TBBPA WITH THE SAME

COLOUR EXCEPT BLACK, WHICH SYMBOLISES THE STANDARD IN PURE SOLVENT. THE PICTURE REPRESENTS THE RESIDUE OF FAT

OBSERVED IN THE FINAL EXTRACT WITH THE THREE VOLUMES OF MILK...... 147

FIGURE 3.15: THREE PROTOCOLS TESTED TO PREPARE HUMAN MILK BASED ON PROTEINS AND LIPIDS REMOVAL...... 150

FIGURE 3.16: TOTAL ION CURRENTS OBTAINED FOR HUMAN MILK TREATED BY SAMPLE PREPARATION METHOD A (RED), B (BLUE) AND

C (GREEN), AND A PROCEDURAL BLANK (BLACK) ANALYSED IN LC-ESI(-)-HRMS...... 152

FIGURE 3.17: MEAN RECOVERY (N=3) OBSERVED FOR SOME OF CL/BR-PHENOLIC COMPOUNDS OF THE STANDARD MIX 1 AFTER

APPLICATION OF THE SAMPLE PREPARATION METHOD A, B AND C AND DETECTION BY LC-ESI(-)-HRMS...... 154

FIGURE 3.18: TOTAL ION CURRENT OBTAINED FOR 400 µL AND 500 µL HUMAN MILK TREATED WITH THE CAPTIVA EMR-LIPID®

PROTOCOL AND CORRESPONDING PROCEDURAL BLANKS ANALYSED BY GC-EI-HRMS...... 157

FIGURE 3.19: MEAN RECOVERY (TOP) AND MATRIX EFFECT (BOTTOM) (N=3) OBSERVED FOR QA/QC MIX 2 COMPOUNDS SPIKED IN

400 µL (RED) AND 500 µL (BLUE) OF HUMAN MILK TREATED WITH THE CAPTIVA EMR-LIPID® APPROACH ANALYSED IN LC- ESI(-)-HRMS...... 158

FIGURE 3.20: MECONIUM (500 MG) SAMPLE PREPARATION WITH 3 EXTRACTIONS APPROACHES E1: 400 µL WATER + 1.6 ML

ACETONITRILE; E2: 800 µL WATER + 3.2 ML ACETONITRILE; E3: 1.6 ML ACETONITRILE AND PURIFIED WITH THE CAPTIVA

EMR-LIPID® CARTRIDGE...... 160

FIGURE 3.21: TOTAL ION CURRENTS OBTAINED FOR 200 MG (RED), 400 MG (BLUE) AND 500 MG (BLACK) OF MECONIUM PREPARED

WITH THE CAPTIVA EMR-LIPID® PROTOCOL AND ANALYSED BY LC-ESI(-)-HRMS...... 162

FIGURE 3.22: REPEATABILITY OF EXTRACTIONS FOR 200 MG (GREY), 400 MG (ORANGE) AND 500 MG (RED) OF MECONIUM TREATED

WITH THE CAPTIVA EMR-LIPID® PROTOCOL AND ANALYSED BY LC-ESI(+/-)-HMRS AND GC-EI-HRMS...... 163

FIGURE 3.23: MEAN RECOVERIES OBSERVED (N=3) FOR 200 MG (GREY), 400 MG (ORANGE) AND 500 MG (RED) OF MECONIUM

TREATED WITH THE SAME PROTOCOL AND ANALYSED BY LC-ESI(+/-)-HMRS AND GC-EI-HRMS. DIAGRAM BARS REPRESENT

THE RECOVERY AFTER THE FIRST ELUTION AND POSITIVE ERROR BARS THE GAIN WITH THE SECOND ELUTION...... 164

- 256/296 - List of figures

FIGURE 3.24: LIQUID-LIQUID PARTITIONING OF THE QA/QC MIX 2 COMPOUNDS BETWEEN POLAR (ACETONITRILE/EAU IN BLUE) AND

NONPOLAR (HEXANE IN RED) PHASES DETECTED IN LC-ESI(+/-)-HRMS (TOP AND MIDDLE LEFT) AND GC-EI-HRMS (MIDDLE

RIGHT AND BOTTOM). *: COMPOUNDS DETECTED IN GC BUT PREFERABLY ANALYSED IN LC...... 166

FIGURE 3.25: RESULTS OF THE OPLS MODEL BUILT TO PREDICT RECOVERY FROM A SET OF PHYSICOCHEMICAL PROPERTIES OF THE

CONSIDERED BIOMARKERS OF EXPOSURE. COMPOUNDS 31, 32 AND 33 WERE TEST COMPOUNDS TO ASSESS MODEL ACCURACY.

EQUATION OF THE LINEAR REGRESSION CURVE IS Y = 0.9942X + 0.0006; R2 = 0.77; RMSEMIN = 0.087 AND RMSEMAX =

0.118 ARE REPRESENTED AS UPPER AND LOWER LIMITS IN BLACK DOTTED LINE AND GREY DASH LINE, RESPECTIVELY...... 175

FIGURE 3.26: CONCENTRATION (NG G -1 LIPID) OF ORGANIC CONTAMINANTS DETECTED IN HUMAN MILK SAMPLES PREPARED WITH THE

TARGETED METHOD DEVELOPED BY LABERCA AND ANALYSED IN GC-EI-MAGNETIC SECTOR, LC-ESI-QQQ, LC-ESI-ORBITRAP,

APGC-QQQ...... 179

FIGURE 3.27: FLUORESCENCE RESPONSE OF WHOLE EXTRACT BIOASSAY OF HUMAN MILK PREPARED WITH THE CAPTIVA EMR-LIPID®

PROTOCOL AND SPIKED WITH 1 µL (BLUE TRIANGLE) AND 4 µL (RED TRIANGLE) OF THE VU MIX AND COMPARED TO

PROCEDURAL BLANK (BLUE DOT) AND UNSPIKED MILK (BLACK RECTANGLE). THE SAME SAMPLE WAS SPIKED WITH 50 µL

(ORANGE DIAMOND), 200 µL (GREY DIAMOND) AND 400 µL (GREEN DIAMOND) OF STANDARDS OF THE QA/QC MIX 2. ... 183

FIGURE 3.28: RESULTS OF THE PERFORMED EDA ASSAY WITH THE BIOLOGICAL RESPONSE DETECTED BY FLUORESCENCE (TOP) FOR

SPIKED (1 µL IN BLUE AND 4 µL IN RED) AND UNSPIKED MILK (GREY) OVERLAPPED WITH THE CORRESPONDING EXTRACTED ION

CHROMATOGRAMS OBTAINED FOR STANDARDS DETECTED BY LC-ESI(-)-HRMS...... 184

FIGURE 3.29: RESULTS OF THE PERFORMED EDA ASSAY WITH THE BIOLOGICAL RESPONSE DETECTED BY FLUORESCENCE (TOP) FOR

HUMAN MILK SPIKED WITH 50 µL (TOP), 200 µL (MIDDLE) AND 400 µL (BOTTOM) STANDARDS OF THE QA/QC MIX 2

OVERLAPPED WITH THE CORRESPONDING EXTRACTED ION CHROMATOGRAMS OBTAINED FOR STANDARDS DETECTED BY LC-ESI(- )-HRMS...... 186

FIGURE 4.1: MD PLOT OBTAINED WITH HALOSEEKER V1.0 SOFTWARE FOLLOWING THE DATA PROCESSING OF A PROCEDURAL BLANK

(TOP), 1 G HUMAN ADIPOSE TISSUE (MIDDLE) AND 0.5 G FAT FROM MUSSEL (BOTTOM) SAMPLES PURIFIED BY ACIDIC

HYDROLYSIS AND ANALYSED BY LC-ESI(-)-HRMS UNDER THE SAME CONDITIONS...... 196

FIGURE 4.2: OVERLAP OF TOTAL ION CURRENTS OBTAINED FOR THE CHLOROFORM FRACTION OF PLACENTA (BLUE) AND PROCEDURAL

BLANK (BLACK) SAMPLES PURIFIED WITH A BLIGH AND DYER EXTRACTION AND ANALYSED BY LC-ESI(-)-HRMS...... 198

FIGURE 4.3: MD PLOT OBTAINED FROM HALOSEEKER V1.0 SOFTWARE FOR A PLACENTA SAMPLE PURIFIED WITH A BLIGH AND DYER

EXTRACTION AND ANALYSED BY LC-ESI(-)-HRMS, WITH A ZOOM ON THE CLUSTER ANNOTATED AS TRICLOSAN...... 199

FIGURE 4.4: FRAGMENTATION MASS SPECTRA OF TRICLOSAN (PRECURSOR ION 286.94388 M/Z) DETECTED BY LC-ESI(-)-HRMS FOR

LOW (NCE 10%) AND HIGH (NCE 60%) COLLISION ENERGIES (TOP AND BOTTOM, RESPECTIVELY), WITH THE ANNOTATION OF

THE OBSERVED DIAGNOSTIC FRAGMENTS...... 200

FIGURE 4.5: OVERLAP OF TOTAL ION CURRENTS OBTAINED FOR THE POLAR FRACTION OF HUMAN MILK (BLUE) AND PROCEDURAL BLANK

(BLACK) SAMPLES PURIFIED WITH THE CAPTIVA EMR-LIPID® PROTOCOL AND ANALYSED BY LC-ESI(-)- AND ESI(+)-HRMS, TOP

AND BOTTOM RESPECTIVELY...... 203

FIGURE 4.6: IDENTIFICATION OF AN UNKNOWN HALOGENATED CLUSTER DETECTED HUMAN MILK PURIFIED WITH THE CAPTVA EMR-

LIPID® PROTOCOL AND ANALYSED BY LC-ESI(-)-HRMS. (A1) MASS DEFECT PLOT GENERATED BY HALOSEEKER V1.0 WITH

POLYHALOGENATED CLUSTERS DETECTED IN UNSPIKED SAMPLES ONLY. THE HALOGENATED CLUSTER AT M/Z 244 IS ZOOMED IN

THE RECTANGLE. (A2) EXPERIMENTAL MASS SPECTRUM OF THIS CLUSTER AND, IN MIRROR, THE THEORETICAL MASS SPECTRUM

- 257/296 - List of figures

OF HYDROXY-CHLOROTHALONIL (C8HCL3N2O) PROPOSED BY HALOSEEKER V1.0. (B) FRAGMENTATION MASS SPECTRA AT NCE

60% OF THE UNKNOWN CLUSTER (TOP) AND OF THE PURE ANALYTICAL STANDARD OF 4-HYDROXYCHLOROTHALONIL (BOTTOM)...... 204

FIGURE 4.7: EXTRACTED ION CHROMATOGRAM OF 4-HYDROXYCHLOROTHALONIL DETECTED IN HUMAN MILK SAMPLES PURIFIED WITH

THE CAPTIVA EMR-LIPID® PROTOCOL AND ANALYSED BY LC-ESI(-)-HRMS (LEFT TOP) WITH THE CORRESPONDING THREE

SCANS FOR THE FRAGMENTATION OF THE PRECURSOR ION (M/Z 244.98) WITH THE IDDA APPROACH (LEFT BOTTOM), AND THE

RESULTING SPECTRA OF FRAGMENTATION (NCE 10, 30 AND 60%) (RIGHT)...... 205

FIGURE 4.8: OVERLAP OF TOTAL ION CURRENTS OBTAINED FOR THE NONPOLAR FRACTION OF HUMAN MILK (BLUE) AND PROCEDURAL

BLANK (BLACK) SAMPLES PURIFIED WITH THE CAPTIVA EMR-LIPID® PROTOCOL AND ANALYSED BY GC-EI-HRMS...... 207

FIGURE 4.9: IDENTIFICATION OF P,P'-DDE (A) AND HCB (B) IN HUMAN MILK (TOP) BY GC-EI-HRMS WITH COMPARISON TO SPECTRA

FRAGMENTATION OF ANALYTICAL STANDARD (BOTTOM). PEAK LIST WITH MASS DEVIATION IN PPM IN BRACKETS FOR P,P’-DDE:

317.93441 (0.03); 280.96852 (0.5); 245.99962 (0.3); 210.02285 (0.5) AND FOR HCB: 283.80967 (-1.0);

248.84095 (-0.1)...... 208

FIGURE 4.10: OVERLAP OF TOTAL ION CURRENTS OBTAINED FOR MECONIUM (BLUE) AND PROCEDURAL BLANK (BLACK) SAMPLES

PURIFIED WITH THE CAPTIVA EMR-LIPID® PROTOCOL AND ANALYSED BY LC-ESI(-)- AND ESI(+)-HRMS, TOP AND BOTTOM

RESPECTIVELY...... 210

FIGURE 4.11: OVERLAP OF TOTAL ION CURRENTS OBTAINED FOR MECONIUM (BLUE) AND PROCEDURAL BLANK (BLACK) SAMPLES

PURIFIED WITH THE CAPTIVA EMR-LIPID® PROTOCOL AND ANALYSED BY GC-EI-HRMS...... 211

FIGURE 4.12: FRAGMENTATION MASS SPECTRA OF P,P’-DDE IN MECONIUM SAMPLES (TOP) AND PURE ANALYTICAL STANDARD

(BOTTOM) ANALYSED BY GC-EI-HRMS...... 212

- 258/296 -

LIST OF TABLES

- 259/296 -

List of tables

TABLE 2.1: LIST OF REFERENCE STANDARDS USED FOR METHOD DEVELOPMENT AND PERFORMANCE (FURTHER NAMED

“CL/BR-PHENOLIC COMPOUNDS” CONSTITUTING THE “STANDARD MIX 1”)...... 76

TABLE 2.2: LIST OF THE REFERENCE STANDARDS WHICH SUBSTITUTED THE STANDARD MIX 1 WITH CL/BR-PHENOLIC COMPOUNDS AND

USED FOR METHOD DEVELOPMENT AND PERFORMANCE ASSESSMENT (FURTHER NAMED “QA/QC MIX 2”)...... 78

TABLE 2.3: LC SETTINGS SELECTED AFTER METHOD OPTIMISATION...... 86

TABLE 2.4: ESI SOURCE SETTINGS SELECTED AFTER METHOD OPTIMISATION ...... 89

TABLE 2.5: HRMS SETTINGS SELECTED AFTER METHOD OPTIMISATION...... 89

TABLE 2.6: GC SETTINGS SELECTED AFTER METHOD OPTIMISATION...... 94

TABLE 2.7: HRMS SETTINGS SELECTED AFTER METHOD OPTIMISATION...... 94

TABLE 2.8: INSTRUMENTAL LINEARITY (COEFFICIENT OF DETERMINATION R2) ON A CALIBRATION CURVE FROM 0.001 TO 0.5 NG µL-1

FOR THE QA/QC MIX 2 ANALYSED BOTH IN LC-ESI(+/-) AND GC-EI...... 95

TABLE 2.9: INSTRUMENTAL LIMITS OF DETECTION OBTAINED IN LC-ESI(+/-) AND GC-EI FOR COMPOUNDS OF THE QA/QC MIX 2. . 96

TABLE 2.10: REPEATABILITY (RELATIVE STANDARD DEVIATION) OBSERVED IN LC-ESI(+/-) AND GC-EI FOR COMPOUNDS OF THE

QA/QC MIX 2 AT THE CONCENTRATION 0.1 NG µL-1...... 98

TABLE 2.11: EXACT MASSES OF CHLORINATED AND BROMINATED ISOTOPES...... 104

TABLE 2.12: HALOSEEKER FILTER RULES TO PRIORITISE SIGNALS OF INTEREST FROM PEAK-PICKED FEATURES TO POLYHALOGENATED

CHEMICAL SPECIES...... 105

TABLE 3.1: LIST OF NBFR COMPOUNDS DETECTABLE WITH THE TARGETED METHOD USED TO CHARACTERISE PLACENTA SAMPLES. .. 143

TABLE 3.2: REPEATABILITY (RSD IN %) OBSERVED FOR CL/BR-PHENOLIC COMPOUNDS SPIKED IN 100 µL OF HUMAN MILK PREPARED

WITH METHOD A, B AND C AND ANALYSED BY LC-ESI(-)-HRMS. FRAC: FRACTION; ND: NOT DETECTED...... 153

TABLE 3.3: SAMPLE PREPARATION LINEARITY AND MATRIX EFFECT (ME) IN GC-EI-HRMS AND LC-ESI(+/-)-HRMS, AND TOTAL

RECOVERY AT FOUR CONCENTRATION LEVELS (LEVEL 1 TO 4 = 6.25; 12.5; 62.5; 125 PG µL-1). ND: NOT DETECTED...... 170

TABLE 3.4: SIGNAL VARIABILITY (RELATIVE STANDARD DEVIATION (RSD) OBSERVED FOR ALL COMPOUNDS IN TRIPLICATE IN DIFFERENT

DETECTION MODES, SPIKED BEFORE THE EXTRACTION AT FOUR CONCENTRATION LEVELS (LEVEL 1 TO 4 = 6.25; 12.5; 62.5; 125

PG µL-1) AND SPIKED AFTER EXTRACTION (SPIKED AFTER EXT.) AT 62.5 PG µL-1. *RSD ON DUPLICATE SAMPLE...... 171

TABLE 3.5: SUBSET GROUP OF REFERENCE TEST COMPOUNDS WITH VALIDATED CRITERIA. LOD: LIMIT OF DETECTION IN NG µL-1, ME:

MATRIX EFFECT...... 173

TABLE 3.6: OPLS MODEL OBSERVATIONS WITH EXAMPLES ON RELATED DATA. VIP IS VARIABLE IMPORTANCE IN PROJECTION...... 177

TABLE 3.7: LIST OF STANDARDS (THYROID HORMONE SYSTEM DISRUPTORS) PROPOSED BY VU TO SPIKE SAMPLES FOR EDA WITH THE

CONCENTRATION IN THE MIX IN NG µL-1...... 181

TABLE 3.8: SUMMARY OF DEVELOPED SAMPLE PREPARATION FOR THE NON-TARGETED CHARACTERISATION OF HUMAN SAMPLES,

INCLUDING COMPARISON OF THEIR CAPABILITIES...... 190

- 261/296 -

LIST OF SCIENTIFIC VALORISATIONS

- 263/296 -

List of scientific valorisations

 Articles in Peer-Reviewed scientific journal

Mariane Pourchet, Laurent Debrauwer, Jana Klanova, Elliott J. Price, Adrian Covaci, Noelia Caballero-Casero, Herbert Oberacher, Marja Lamoree, Annelaure Damont, François Fenaille, Jelle Vlaanderen, Jeroen Meijer, Martin Krauss, Dimosthenis Sarigiannis, Robert Barouki, Bruno Le Bizec, Jean-Philippe Antignac, Suspect and non-targeted screening of chemicals of emerging concern for human biomonitoring, environmental health studies and support to risk assessment: From promises to challenges and harmonisation issues, Environment International 2020, 139, 105545.

Mariane Pourchet, Luca Narduzzi, Annabelle Jean, Ingrid Guiffard, Emmanuelle Bichon, Ronan Cariou, Yann Guitton, Sébastien Hutinet, Jelle Vlaanderen, Jeroen Meijer, Bruno Le Bizec, Jean-Philippe Antignac, Non-targeted screening methodology to characterise human internal chemical exposure: application to halogenated compounds in human milk, Under review in Talanta

 Proceeding and extended abstracts

Mariane Pourchet, Ronan Cariou, Jean-Philippe Antignac, Bruno Le Bizec Screening emerging chemicals in human matrices to support biomonitoring and environmental health studies: methods, challenges and promises, Dioxin2018 – Krakow (Poland) – 26-31 August 2018

Mariane Pourchet, Ronan Cariou, Emmanuelle Bichon, Bruno Le Bizec, Jean-Philippe Antignac, Non-targeted screening of markers of chemical exposure in human breast milk: development and challenges regarding sample preparation and data processing, SETAC SciCon 2020 – Dublin (Ireland) – 3-7 May 2020 (virtual conference)

 Published reports o 2017

Mariane Pourchet, Jean-Philippe Antignac, Laurent Debrauwer, Adrian Covaci, Screening methods inventory, HBM4EU Deliverable report, 2017

- 265/296 - List of scientific valorisations

o 2018

Mariane Pourchet, Jean-Philippe Antignac, Laurent Debrauwer, Noelia Caballero-Casero, Adrian Covaci, Workflow for screening emerging chemicals, HBM4EU Deliverable report, 2018

o 2020

Jérémie Achille, Tamara Braish, Marc Codaccioni, Hélène Desqueyroux, Philippe Glorennec, Benjamin Hanoune, Marion Hulin, Corinne Mandin, Mélanie Nicolas, Mariane Pourchet, Gaëlle Raffy, Corentin Regrain, Wenjuan Wei, Conférence sur les environnements bâtis, naturels et sociaux 18-23 août 2019, Kaunas (Lituanie), Environnement, Risques & Santé 2020, 19 (1), 51-57

 Oral communications o 2018

Mariane Pourchet, Jean-Philippe Antignac, Emmanuelle Bichon, Ronan Cariou, Bruno Le Bizec, Les défis de la préparation d’échantillon pour l’analyse non ciblée au service de la caractérisation de l’exposome, Journée thématique CCOA – Nantes (France) – 16 October 2018

o 2019

Mariane Pourchet, Ronan Cariou, Emmanuelle Bichon, Sébastien Hutinet1, Bruno Le Bizec, Jean-Philippe Antignac, Non-targeted approaches to screen new markers of chemical exposure from human matrices: application and challenges regarding sample preparation and data processing, ISES/ISEE 2019 – Kaunas (Lithuania) – 18-22 August 2019

Mariane Pourchet, Ronan Cariou, Emmanuelle Bichon, Bruno Le Bizec, Jean-Philippe Antignac, Development and application of a non-targeted approach to characterise human exposure to halogenated chemicals of concern in human breast milk, JBS 2019 - Annual meeting of the doctoral school “Biologie-Santé” – Angers (France) – 10-11 December 2019

o 2020

- 266/296 - List of scientific valorisations

Mariane Pourchet, Ronan Cariou, Emmanuelle Bichon, Bruno Le Bizec, Jean-Philippe Antignac, Non-targeted screening of markers of chemical exposure in human breast milk: development and challenges regarding sample preparation and data processing, SETAC SciCon 2020 – Dublin (Ireland) – 3-7 May 2020 (virtual conference)

 Posters

Mariane Pourchet, Ronan Cariou, Jean-Philippe Antignac, Bruno Le Bizec, Screening emerging chemicals in human matrices to support biomonitoring and environmental health studies: methods, challenges and promises, Dioxin2018 – Krakow (Poland) – 26 – 31 August 2018

Mariane Pourchet, Emmanuelle Bichon, Ronan Cariou, Bruno Le Bizec, Jean-Philippe Antignac, Les défis de l’analyse non-ciblée appliquée à la caractérisation de l’exposome interne chez l’Homme, SEP2019 – Paris (France), Flash-poster

 Conference attending

Séminaire Agreenium « GlobalHealth – Biodiversité, santé des écosystèmes, santé humaine » – Lyon (France) – 26-30 March 2020

Journée thématique CCOA « Fondamentaux de la chromatographie et troubleshooting » – Nantes (France) – 13 June 2019

Journée thématique CCOA « Les sciences séparatives appliquées à l'analyse des macromolécules » – Nantes (France) – 10 March 2020

- 267/296 - List of scientific valorisations

 HBM4EU meetings o Oral presentations

Mariane Pourchet, Jean-Philippe Antignac, Screening emerging chemicals in human matrices to support biomonitoring and environmental health studies: focus on halogenated compounds, HBM4EU WP16 Working Meeting- Nantes (France) – 6-7 June 2018

Mariane Pourchet, Jean-Philippe Antignac, Action 3 | Untargeted screening of halogenated marker, HBM4EU WP16 Working Meeting- Paris (France) – 20-21 May 2019

Mariane Pourchet, Jean-Philippe Antignac, Action 4 | WP14-WP16 interaction, HBM4EU WP16 Working Meeting- Paris (France) – 20-21 May 2019

o Poster

Mariane Pourchet, Jean-Philippe Antignac, Development and application of non-targeted approaches for characterizing human internal exposure to halogenated chemicals of concern, Consortium meeting – Berlin (Germany) – 9 October 2019; Award for the best poster

 Supervising and training

Supervising of Annabelle Jean during her master’s degree internship, Développement d’un protocole de préparation d’échantillon applicable au lait maternel pour la recherche non ciblée de marqueurs d’exposition chlorés et bromés par LC-HRMS, 7 January – 21 June 2019

Supervising of Lucie Guitton during her master’s degree internship, Développement d’un protocole de préparation d’échantillon et de mesure par couplages LC-HRMS et GC-HRMS pour la recherche non-ciblée de biomarqueurs d’exposition halogénés dans les matrices humaines, 9 March – 4 September 2020

Training of Fernando Vela-Soria (postdoctoral fellow from the University of Granada) to non- targeted strategies (sample preparation, LC- and GC-HRMS analysis and data processing) in order to characterise placenta samples. Collaboration in the frame of the HBM4EU project, under the WP14-WP16 interaction, March – September 2020.

- 268/296 -

APPENDICES

- 269/296 -

Appendices

Appendix 1: Recent proposition of confidence level scale for the annotation of metabolites proposed by the Metabolite Identification Task Group of Metabolomics Society. It includes additional information on the compound structure compared to the scale proposed by Schymanski et al., 2015, such as the chirality and stereochemistry of the molecule. At the present time, no consensus has been adopted and the Metabolite Identification Task Group was created for dealing with this issue.

Reference:http://metabolomicssociety.org/images/newsletters/Revised%20Structure- Based%20Reporting%20Standards%20for%20Metabolite%20Identification_141119.pdf

- 271/296 - Appendices

Appendix 2: Chemical structure of compounds included in the QA/QC mix and list of their InChIKey.

- 272/296 - Appendices

- 273/296 - Appendices

- 274/296 - Appendices

No. Component InChIKey 1 2.4-Dichlorophenol HFZWRUODUSTPEG-UHFFFAOYSA-N 2 3.5.6-Trichloro-2-pyridinol WCYYAQFQZQEUEN-UHFFFAOYSA-N 3 Simazine ODCWYMIRDDJXKW-UHFFFAOYSA-N 4 Fenvalerate free acid VTJMSIIXXKNIDJ-UHFFFAOYSA-N 5 2.3.4.5-Tetrachlorophenol RULKYXXCCZZKDZ-UHFFFAOYSA-N 6 2.4-Dibromophenol FAXWFCTVSHEODL-UHFFFAOYSA-N 7 Acetochlor VTNQPKFIQCLBDU-UHFFFAOYSA-N 8 Hexaclorobenzene CKAPSXZOOQJIBF-UHFFFAOYSA-N 9 Metolachlor WVQBLGZPHOPPFO-UHFFFAOYSA-N 10 beta-Hexachloroclyclohexane JLYXXMFPNIAWKQ-UHFFFAOYSA-N 11 Triclosan XEFQLINVKFYRCS-UHFFFAOYSA-N 12 Fenhexamid VDLGAVXLJYLFDH-UHFFFAOYSA-N 13 p.p'-dichlorodiphenyldichloroethylene UCNVFOCBFJOQAL-UHFFFAOYSA-N 14 Chlorpyrifos SBPBAQFWLVIOKP-UHFFFAOYSA-N 15 Chlorfenvinphos FSAVDKDHPDSCTO-XYOKQWHBSA-N 16 Tetraconazole LQDARGUHUSPFNL-UHFFFAOYSA-N 17 Quizalofop-ethyl OSUHJPCHFDQAIT-GFCCVEGCSA-N 18 Prochloraz TVLSRXXIMLFWEO-UHFFFAOYSA-N 19 (Z)-Dimethomorph QNBTYORWCCMPQP-JXAWBTAJSA-N 20 2.3.4.5-Tetrabromophenol KXZRECGEDVBJPM-UHFFFAOYSA-N 21 2.3.5.6-tetrabromo-p-xylene RXKOKVQKECXYOT-UHFFFAOYSA-N 22 Fipronil ZOCSXAVNDGMNBV-UHFFFAOYSA-N 23 Deltamethrin OWZREIFADZCYQD-NSHGMRRFSA-N 24 Tetrabromobisphenol A VEORPZCZECFIRK-UHFFFAOYSA-N 25 Hexabromobenzene CAYGQBVSOZLICD-UHFFFAOYSA-N 26 alpha-hexabromocyclododecane DEIGXXQKDWULML-PQTSNVLCSA-N 27 Pentabromodiphenyl ether 153 RZXIRSKYBISPGF-UHFFFAOYSA-N 28 anti-Dechlorane plus UGQQAJOWXNCOPY-VBCJEVMVSA-N 6-Hydroxy-2.2.3.4.4.5-hexabromodiphenyl 29 LDMKXEGTHGJWLG-UHFFFAOYSA-N ether 30 1.2-Bis(2.4.6-tribromophenoxy)ethane YATIGPZCMOYEGE-UHFFFAOYSA-N

- 275/296 - Appendices

Appendix 3: Error in the cluster alignment due to shift of peak intensity.

- 276/296 - Appendices

Appendix 4: LC-MS-MS/MS method to generate fragment mass spectra at normalised collision energy 10, 20, 30, 40, 50 and 60.

 LC settings

Stationary phase Hypersil Gold (50 mm × 2.1 mm, 1.9 μm)

Mobile phase (A) Water + 10 mM ammonium acetate (B) Acetonitrile + 10 mM ammonium acetate

Gradient

Solvent flow 0.6 mL/min

Oven temperature 45 °C

 ESI source settings

Heater temperature 350 °C

Capillary temperature 350 °C

Sheath gas flow 50 AU

Auxiliary gas flow 10 AU

Sweep gas flow 0 AU

S-lens radio frequency 70 AU

3.5 kV positive mode Spray voltages -2.5 kV negative mode

- 277/296 - Appendices

 MS settings

Full-scan MS/MS

Resolution 35,000 FWHM at m/z 200 17,500 FWHM at m/z 200

Mass-to-charge (m/z) range 100-1 000 1st mas fixed at 50

AGC** target 5×105 1×105

Maximum injection time 250 ms 80 ms

Ion mode Positive or negative

NCE / 10, 20, 30, 40, 50, 60

Isolation window / 1.0 m/z

- 278/296 - Appendices

Appendix 5: Tutorial “How to create a GC-HRMS library with MS-DIAL”, written by Noelia Caballero Casero and Mariane Pourchet

How to create a GC-HRMS library with MS-DIAL

Written by Noelia Caballero Casero and Mariane Pourchet

Version 2: 17th March 2020

- Convert data with MSConvert into centroid data in mzML format - Open MS-DIAL software - To create a new project, follow MS-DIAL tutorial section 4 “start up a project of MS- DIAL” - Open the project on the format “mtd2” - Double click on the sample of interest or on the alignment result file to visualise the plot (the triangle means the peak was deconvoluted) - All compounds are referenced with a compound ID which can be used for querying the NIST

- 279/296 - Appendices

- To export the peak list, click “export- peak list result”, select where the peak list will be saved, choose the file of interest and click “add”, select export format “MSP”, and click “export”

- Open NIST software “MSSEARCH-nistms (application)” (if NIST is opened for the first time, choose the database directory to the MSSEARCH folder). - Import the peak list “file-open”, choose the peak list and import all spectra or selected those of interest regarding the compound ID.

- Select the compound of interest regarding its ID in the scrolling menu and manually verify the identity of the compound.

- 280/296 - Appendices

- Since the compound has been identified, click on InChIKey to access PubChem and get all required information to fill the library. - Back to MS-DIAL, there are two options: 1. Click on the deconvoluted spectra. Then, open MS-FINDER with this button

2. Right click on “peak spot viewer”, Search formula and structure < Go to MS- FINDER program < Deconvoluted.

- 281/296 - Appendices

- If the “file navigator” is full of unwanted files, go to MSFINDER folder, in MSDIAL_TEMP and delete all previous files - Fill all “file information” - To add fragment identification to the final library file, click “Analysis-Fragment annotation” - Export the file to a MSP format “export-export peak annotation result as MSP format”

- 282/296 - Appendices

- The MSP file can be opened with Notepad to manually modify the exact mass of the [M+] precursor ion and the RI - Individual MSP file can be combined with TXT Collector - To re-run sample with the new library from the identification step, go to MS-DIAL “Data processing-identification” and load the new library in MSP file.

- 283/296 - Appendices

Appendix 6: Concentrations of novel brominated flame retardants in ng/g fat in human milk and placenta after analysis with a targeted approach.

PBDE 28 PBDE 47 PBDE 49 PBDE 99 PBDE 100 PBDE 153 PBDE 154 PBDE 183 PBDE 209 PBB 52 PBB 101 PBB 153 Milk 1 0.030 0.400 0.030 0.287 0.131 0.519 0.040 0.061 1.411 ND ND 0.060 Milk 2 0.415 6.080 0.075 1.172 0.791 1.035 0.093 0.137 0.820 ND ND 0.056 Milk 3 0.041 0.371 0.007 0.074 0.098 0.385 0.011 0.019 0.214 ND ND 0.120 Placenta 1 0.013 0.125 ND 0.027 0.013 0.126 0.007 0.020 0.521 ND ND 0.014 Placenta 2 0.013 0.125 ND 0.029 0.017 0.127 ND 0.023 0.392 ND ND 0.019 Placenta 3 0.010 0.187 ND 0.032 0.022 0.188 ND 0.017 0.417 ND ND 0.010 Placenta 4 0.014 0.176 ND 0.024 0.038 0.075 ND 0.016 10.405 ND ND ND Placenta 5 0.010 0.178 ND 0.037 0.054 0.246 ND 0.035 0.630 ND ND ND Placenta 6 0.015 0.071 ND 0.027 0.025 0.109 ND ND 3.224 ND ND 0.025

PBEB PBT TBCT nHBB nPBB pTBX OBIND DBDPE T23BPIC a-HBCDD b-HBCDD g-HBCDD Milk 1 ND ND ND 0.142 ND ND ND 1.961 ND 0.756 ND ND Milk 2 ND ND ND 0.049 ND ND ND 1.251 ND 0.413 ND ND Milk 3 ND ND ND 0.058 ND ND ND 1.182 ND 0.474 ND ND Placenta 1 ND ND ND 0.110 ND ND ND ND ND ND ND ND Placenta 2 ND ND ND 0.142 ND ND ND 3.422 ND ND ND ND Placenta 3 ND ND ND 0.147 ND ND ND 2.952 ND ND ND ND Placenta 4 ND ND ND 0.139 ND ND ND 6.566 ND ND ND ND Placenta 5 ND ND ND 0.114 ND ND ND 5.567 ND ND ND ND Placenta 6 ND ND ND 0.116 ND ND ND 77.150 ND ND ND ND

- 284/296 - Appendices

TBBPA TBBPS MonoBP DiBP TriBP TetraBP PentaBP Milk 1 0.43 ND ND ND 0.15 ND ND Milk 2 0.04 ND ND ND 0.06 ND ND Milk 3 0.21 ND ND ND 0.06 ND ND Placenta 1 0.44 ND ND ND 0.39 ND ND Placenta 2 0.64 ND 8.88 ND 0.65 0.04 ND Placenta 3 0.26 ND ND ND 0.42 0.02 ND Placenta 4 0.21 ND 4.43 ND 0.29 0.01 ND Placenta 5 0.55 ND ND ND 0.36 0.03 ND Placenta 6 1.06 ND ND ND 0.29 0.02 ND

EHTBB BEHTBP TBBPA bME BTBPE TBBPA-bDiBPrE Milk 1 ND ND ND ND ND Milk 2 ND ND ND ND ND Milk 3 ND ND ND ND ND Placenta 1 ND ND ND ND ND Placenta 2 ND ND ND ND ND Placenta 3 ND ND ND ND ND Placenta 4 ND ND ND ND ND Placenta 5 ND ND ND ND ND Placenta 6 ND ND ND ND ND

- 285/296 - Appendices

Appendix 7: Sample preparation protocol of human milk and meconium based on the Captiva EMR-Lipid® cartridge

Human milk

 Vigorously shake human milk at room temperature  Pipet 400 µL of milk in a 6 mL test tube  Next steps are common for human milk and meconium

Meconium

 Weigh 500 mg of meconium in a 6 mL test tube  Centrifuge for 10 min, 3000 rpm, 4 °C  Add 400 µL of water  Vortex for at least 30 sec and make sure the sample is partly dissolved  Next steps are common for human milk and meconium

Common protocol for both human milk and meconium

 Add internal standard previously diluted in acetonitrile  Add a total volume of 1.6 mL of acetonitrile taking into account the volume of internal standard previously added  Vortex for 30 sec  Centrifuge for 10 min, 3000 rpm, 4 °C  While of conditioning cartridge, store samples at 4°C  Rinse Captiva EMR-Lipid (6 mL, 600 mg) with 2*5 mL acetonitrile/water 8:2 (v/v)  Make sure the top of the stationary phase has not dried  Load the sample supernatant into the cartridge  Eluate at atmospheric pressure in a 15 mL test tube  Extract the polar fraction with 2*2 mL of hexane  Collect nonpolar fractions in a 15 mL test tube  Remove water residues from hexane with Na2SO4  Transfer the nonpolar fraction in a 6 mL test tube with taking Na2SO4

- 286/296 - Appendices

 Concentrate under a gentle stream of nitrogen both fractions until 100 µL, at 35 °C for the polar fraction and at room temperature for the nonpolar fraction  Transfer fractions in a vial with a 200 µL insert  Rinse test tube with 100 µL of acetonitrile and hexane for polar and nonpolar fraction respectively and vortex  Evaporate both fractions under a gentle stream of nitrogen until dryness, at 35 °C for the polar fraction and at room temperature for the nonpolar fraction  Reconstitute polar fraction in 50 µL of acetonitrile/water 8:2 (v/v) containing external standard  Reconstitute nonpolar fraction in 50 µL of hexane containing external standard  Store sample at 4°C for 15 min  Centrifuge for 5 min, 300 rpm at room temperature  If there is a residue in the vial, transfer the extract in a new vial  Analyse polar fraction with a LC-HRMS system  Analyse nonpolar fraction with a GC-HRMS system

- 287/296 - Appendices

Appendix 8: Information extracted from PubChem for each compound of the QA/QC mix 2 to build the OPLS model

Hydrogen Hydrogen Bond Rotatable Exact No. Name XLogP3 Bond Donor Acceptor Count Bond Count Mass Count

1 2.4-DCP 3,1 1 1 0 161,964 2 TCPy 3,2 1 1 0 196,92 3 Simazine 2,2 2 5 4 201,078 4 Fenvalerate free acid 3,4 1 2 3 212,06 5 2.3.4.5-tetraCP 4,5 1 1 0 231,883 6 2.4-DBP 3,2 1 1 0 251,861 7 Acetochlor 3,2 0 2 6 269,118 8 HCB 5,7 0 0 0 283,81 9 Metolachlor 3,1 0 2 6 283,134 10 β-HCH 3,8 0 0 0 289,857 11 Triclosan 5 1 2 2 287,951 12 Fenhexamid 4,4 2 2 2 301,064 13 p.p'-DDE 7 0 0 2 317,935 14 Chlorpyrifos 5,3 0 5 6 348,926 15 Chlorfenvinphos 3,1 0 4 7 357,97 16 Tetraconazole 4,4 0 7 7 371,022 17 Quizalofop-ethyl 4,3 0 6 7 372,088 18 Prochloraz 4,6 0 3 6 375,031 19 (Z)-Dimethomorph 3,9 0 4 5 387,124 20 2.3.4.5-tetraBP 5,4 1 1 0 409,68 21 p-TBX 5,4 0 0 0 421,716 22 Fipronil 4,5 1 11 2 435,939 23 Deltamethrin 6,2 0 4 7 504,971 24 TBBPA 6,8 2 2 2 543,753 25 HBBz 6,1 0 0 0 551,504 26 α-HBCDD 7,1 0 0 0 641,645 27 PBDE 153 7,6 0 1 2 643,53 28 a-DP 8 0 0 0 653,711 29 OH-BDE 137 7,2 1 2 2 659,525 30 BTBPE 7,7 0 2 5 687,556

- 288/296 - Appendices

Heavy Topological Polar Defined Atom No. nm md RMD Atom Complexity Surface Area Stereocenter Count Count 1 162 -0,03608 -222,716 20,2 9 97,1 0 2 197 -0,079803 -405,091 29,1 10 243 0 3 201 0,078123 388,672 62,7 13 131 0 4 212 0,060407 284,939 37,3 14 196 0 5 232 -0,116975 -504,203 20,2 11 143 0 6 252 -0,13916 -552,222 20,2 9 97,1 0 7 269 0,118257 439,617 29,5 18 260 0 8 284 -0,189834 -668,43 0 12 104 0 9 283 0,133907 473,17 29,5 19 285 0 10 290 -0,142884 -492,703 0 12 104 0 11 288 -0,048837 -169,573 29,5 17 252 0 12 301 0,063634 211,409 49,3 19 331 0 13 318 -0,064939 -204,211 0 18 269 0 14 349 -0,073716 -211,221 72,7 18 303 0 15 358 -0,030471 -85,1145 44,8 20 368 0 16 371 0,02153 58,0323 39,9 23 381 0 17 372 0,087685 235,712 70,5 26 459 1 18 375 0,03081 82,16 47,4 23 377 0 19 387 0,123736 319,731 480 27 512 0 20 410 -0,32018 -780,927 20,2 11 143 0 21 422 -0,2838 -672,512 0 12 127 0 22 436 -0,061294 -140,583 1040 26 599 0 23 505 -0,02888 -57,1881 59,3 28 643 3 24 544 -0,24702 -454,081 40,5 21 310 0 25 552 -0,49612 -898,768 0 12 104 0 26 642 -0,35527 -553,38 0 18 175 6 27 644 -0,46991 -729,674 9,2 19 276 0 28 654 -0,288718 -441,465 0 30 782 0 29 660 -0,47499 -719,682 29,5 20 333 0 30 688 -0,44369 -644,898 18,5 22 291 0

- 289/296 - Appendices

Undefined Atom Defined Bond Number of No. C H N O Cl Br P S F Stereocenter Count Stereocenter Count insaturation

1 0 0 6 4 0 1 2 0 0 0 0 4 2 0 0 5 2 1 1 3 0 0 0 0 4 3 0 0 7 12 5 0 1 0 0 0 0 4 4 1 0 11 13 0 2 1 0 0 0 0 5 5 0 0 6 2 0 1 4 0 0 0 0 4 6 0 0 6 4 0 1 0 2 0 0 0 4 7 0 0 14 20 1 2 1 0 0 0 0 5 8 0 0 6 0 0 0 6 0 0 0 0 4 9 1 0 15 22 1 2 1 0 0 0 0 5 10 0 0 6 6 0 0 6 0 0 0 0 1 11 0 0 12 7 0 2 3 0 0 0 0 8 12 0 0 14 17 1 2 2 0 0 0 0 6 13 0 0 14 8 0 0 4 0 0 0 0 9 14 0 0 9 11 1 3 3 0 1 1 0 4 15 0 1 12 14 0 4 3 0 1 0 0 5 16 1 0 13 11 3 1 2 0 0 0 4 7 17 0 0 19 17 2 4 1 0 0 0 0 12 18 0 0 15 16 3 2 3 0 0 0 0 8 19 0 1 21 22 1 4 1 0 0 0 0 11 20 0 0 6 2 0 1 0 4 0 0 0 4 21 0 0 8 6 0 0 0 4 0 0 0 4 22 1 0 12 4 4 1 2 0 0 1 6 9 23 0 0 22 19 1 3 0 2 0 0 0 13 24 0 0 15 12 0 2 0 4 0 0 0 8 25 0 0 6 0 0 0 0 6 0 0 0 4 26 0 0 12 18 0 0 0 6 0 0 0 1 27 0 0 12 4 0 1 0 6 0 0 0 8 28 8 0 18 12 0 0 12 0 0 0 0 7 29 0 0 12 4 0 2 0 6 0 0 0 8 30 0 0 14 8 0 2 0 6 0 0 0 8

- 290/296 - Appendices

Prsence of OH group or Presence of No. Non ramified cycle Recovery oxygen double bond OH group 1 1 1 1 0.000 2 1 0 1 0.362 3 0 0 0 0.326 4 1 1 0 0.251 5 1 1 1 0.246 6 1 1 1 0.019 7 1 0 0 0.394 8 0 0 1 0.089 9 1 0 0 0.393 10 0 0 1 0.073 11 1 1 0 0.416 12 1 1 0 0.440 13 0 0 0 0.360 14 0 0 0 0.538 15 0 0 0 0.284 16 0 0 0 0.390 17 1 0 0 0.645 18 1 0 0 0.170 19 1 0 0 0.333 20 1 1 1 0.316 21 0 0 1 0.019 22 1 0 0 0.390 23 1 0 0 0.384 24 1 1 0 0.399 25 0 0 1 0.088 26 0 0 0 0.292 27 0 0 0 0.035 28 0 0 0 0.004 29 1 1 0 0.265 30 0 0 0 0.003

- 291/296 - Appendices

Appendix 9: LC-HRMS and fractionation method for bioassays developed at VU and used to analyse human milk.

LC-separation was performed on an Agilent ZORBAX Rapid Resolution High Definition (RRHD) Eclipse plus C18 (2.1 x 50 mm, 1.8 µm particle size) column at 30 °C using an Agilent Infinity 1290 UPLC pump and auto sampler. 20 µL of standard, sample or blank (methanol/water 1:4 (v/v)) was injected at a flow rate of 500 µL/min in 90% mobile phase A (100% water) and 10% mobile phase B (100% acetonitrile). The solvent gradient increased to 99% mobile phase B over an 18 minutes period. This was kept for another 12 minutes after which the gradient decreased to 10% mobile phase B and 90% mobile phase A again in 0.5 minute. This ratio was kept until a total runtime of 35 minutes. Spectra’s were recorded and fractions were collected in the first 18 minutes of the run. After LC-separation the flow was diverted to either the Bruker Compact QTOF-MS for chemical analysis or the FractioMate for the collection of fractions to be used in the bioassay.

The ionisation source that was used for chemical analysis was Electron Spray Ionisation (ESI) in negative mode. The endplate voltage and the capillary voltage yielded 500V and 4500V, respectively. The nebuliser pressure was kept at 1,8 bar, the dry gas flow at 9 L/min and the temperature at 220°C. Everything was scanned in single MS mode with a scan range ranging from 50m/z to 1300m/z and a spectra rate of 4 Hz. Before each sample injection, 20uL of sodium formate was directly injected for calibration.

Fractions were collected in black non-binding polystyrene 96-wells plates. Fraction wells were pre-filled with 10 µL of 10% DMSO as keeper to increase recoveries. A totality of 80 fractions were collected covering the wells from A3 to H12 during the first 18 minutes of the run, resulting in 0.23 minute fractions. The remaining wells (A1 to H2) were used for the T4 calibration curve to assess bioassay performance. After fractionation, the plates were dried using the Centrivap concentrator for approximately 4 hours at 25 °C after which the fractions were suspended in bioassay buffer

- 292/296 -

Titre : Développement d'approches de profilage non ciblé pour la mise en évidence de dangers chimiques émergents : Application à l'identification de nouveaux marqueurs d'exposition interne chez l'Homme au service de la biosurveillance et de l'étude du lien entre exposition chimique et santé humaine Mots clés : Analyse non-ciblée, spectrométrie de masse à haute résolution, exposition chimique, biosurveillance, projet européen HBM4EU-H2020 Résumé : L’exposition de l’Homme aux particulier vis-à-vis des contaminants contaminants chimiques présente un risque émergents. Pour pallier ce manque pour sa santé. Des programmes de d’exhaustivité, de nouvelles approches dites biosurveillance sont menés dans de nombreux non-ciblées sont aujourd’hui développées. Le pays pour évaluer à la fois cette exposition, son présent travail de thèse, inscrit dans le projet impact en terme de santé, ainsi que l’efficacité HBM4EU, a eu pour objectif le développement des mesures de gestion et des dispositions de stratégies analytiques non-ciblées depuis la réglementaires mises en place. L’Union préparation d’échantillons, leur profilage par Européenne a lancé en 2017 le projet Human couplage LC- et GC-HRMS, jusqu’au Biomonitoring for Europe (HBM4EU) afin de retraitement des données. Plusieurs matrices consolider un réseau pan-Européen dans ce humaines ont été considérées avec un focus domaine, et de mieux documenter la réalité de autour de l’exposition périnatale. Ce travail a l’exposition chimique des citoyens européens abouti à une première preuve de concept et son impact sur la santé humaine. Les démontrant la capacité de ces approches pour approches historiquement mises en œuvre l’identification de composés émergent sans a pour caractériser l’exposition, de type ciblées, priori. Ces recherches ont également permis sont très performantes pour rechercher et de mettre en évidence les principaux défis quantifier les contaminants déjà connus. En associées à ces approches pour tous revanche, elles ne permettent pas de rendre laboratoires souhaitant rejoindre cette compte de la complexité des expositions, en dynamique.

Title: Development of non-targeted approach to evidence emerging chemical hazards: Identification of new biomarkers of internal human exposure, in order to support human biomonitoring and the study of the link between chemical exposure and human health. Keywords: Non-targeted screening, High resolution mass spectrometry, chemical exposure, biomonitoring, European project HBM4EU-H2020 Abstract: Human exposure to chemical capture new contaminants of emerging contaminants may impact his health. concern. To overcome this lack of Biomonitoring programs are conducted in completeness, new approaches such as many countries to assess both this exposure, suspect and non-targeted screening are today its consequence on human health, and the developed. The present thesis work, part of the efficiency of regulatory and risk management HBM4EU project, aimed to develop such non- provisions. The European Union launched the targeted analytical strategies from sample Human Biomonitoring for Europe (HBM4EU) preparation, LC- and GC-HRMS profiling, to project in 2017 to consolidate a pan-European data processing. Several human matrices were network in this field, and better document the considered with special emphasis on perinatal reality of chemical exposure of European exposure. This work led to a first proof-of- citizens and establish the possible link with concept demonstrating the interest of these human health. The historical approaches used approaches for identifying emerging chemicals to characterise the exposure, known as without a priori. These researches have also targeted methods, are efficient to monitor and highlighted main challenges to be faced in the quantify already known substances. However, development of these approaches for all they do not allow to reflect the complexity of laboratories wishing to join this dynamic. human exposure and in particular do not